The emergence of collaborations, which standardize and combine multiple clinical databases across different regions, provide a wealthy source of data, which is fundamental for clinical prediction models, such as patient-level predictions. With the aid of such large data pools, researchers are able to develop clinical prediction models for improved disease classification, risk assessment, and beyond. To fully utilize this potential, Machine Learning (ML) methods are commonly required to process these large amounts of data on disease-specific patient cohorts. As a consequence, the Observational Health Data Sciences and Informatics (OHDSI) collaborative develops a framework to facilitate the application of ML models for these standardized patient datasets by using the Observational Medical Outcomes Partnership (OMOP) common data model (CDM). In this study, we compare the feasibility of current web-based OHDSI approaches, namely ATLAS and "Patient-level Prediction" (PLP), against a native solution (R based) to conduct such ML-based patient-level prediction analyses in OMOP. This will enable potential users to select the most suitable approach for their investigation. Each of the applied ML solutions was individually utilized to solve the same patient-level prediction task. Both approaches went through an exemplary benchmarking analysis to assess the weaknesses and strengths of the PLP R-Package. In this work, the performance of this package was subsequently compared versus the commonly used native R-package called Machine Learning in R 3 (mlr3), and its sub-packages. The approaches were evaluated on performance, execution time, and ease of model implementation. The results show that the PLP package has shorter execution times, which indicates great scalability, as well as intuitive code implementation, and numerous possibilities for visualization. However, limitations in comparison to native packages were depicted in the implementation of specific ML classifiers (e.g., Lasso), which may result in a decreased performance for real-world prediction problems. The findings here contribute to the overall effort of developing ML-based prediction models on a clinical scale and provide a snapshot for future studies that explicitly aim to develop patient-level prediction models in OMOP CDM.

BACKGROUND: Given the geographical sparsity of Rare Diseases (RDs), assembling a cohort is often a challenging task. Common data models (CDM) can harmonize disparate sources of data that can be the basis of decision support systems and artificial intelligence-based studies, leading to new insights in the field. This work is sought to support the design of large-scale multi-center studies for rare diseases.METHODS: In an interdisciplinary group, we derived a list of elements of RDs in three medical domains (endocrinology, gastroenterology, and pneumonology) according to specialist knowledge and clinical guidelines in an iterative process. We then defined a RDs data structure that matched all our data elements and built Extract, Transform, Load (ETL) processes to transfer the structure to a joint CDM. To ensure interoperability of our developed CDM and its subsequent usage for further RDs domains, we ultimately mapped it to Observational Medical Outcomes Partnership (OMOP) CDM. We then included a fourth domain, hematology, as a proof-of-concept and mapped an acute myeloid leukemia (AML) dataset to the developed CDM.RESULTS: We have developed an OMOP-based rare diseases common data model (RD-CDM) using data elements from the three domains (endocrinology, gastroenterology, and pneumonology) and tested the CDM using data from the hematology domain. The total study cohort included 61,697 patients. After aligning our modules with those of Medical Informatics Initiative (MII) Core Dataset (CDS) modules, we leveraged its ETL process. This facilitated the seamless transfer of demographic information, diagnoses, procedures, laboratory results, and medication modules from our RD-CDM to the OMOP. For the phenotypes and genotypes, we developed a second ETL process. We finally derived lessons learned for customizing our RD-CDM for different RDs.DISCUSSION: This work can serve as a blueprint for other domains as its modularized structure could be extended towards novel data types. An interdisciplinary group of stakeholders that are actively supporting the project's progress is necessary to reach a comprehensive CDM.CONCLUSION: The customized data structure related to our RD-CDM can be used to perform multi-center studies to test data-driven hypotheses on a larger scale and take advantage of the analytical tools offered by the OHDSI community.

BACKGROUND: Machine learning (ML) has emerged as a vital asset for researchers to analyze and extract valuable information from complex datasets. However, developing an effective and robust ML pipeline can present a real challenge, demanding considerable time and effort, thereby impeding research progress. Existing tools in this landscape require a profound understanding of ML principles and programming skills. Furthermore, users are required to engage in the comprehensive configuration of their ML pipeline to obtain optimal performance. RESULTS: To address these challenges, we have developed a novel tool called Machine Learning Made Easy (MLme) that streamlines the use of ML in research, specifically focusing on classification problems at present. By integrating 4 essential functionalities-namely, Data Exploration, AutoML, CustomML, and Visualization-MLme fulfills the diverse requirements of researchers while eliminating the need for extensive coding efforts. To demonstrate the applicability of MLme, we conducted rigorous testing on 6 distinct datasets, each presenting unique characteristics and challenges. Our results consistently showed promising performance across different datasets, reaffirming the versatility and effectiveness of the tool. Additionally, by utilizing MLme's feature selection functionality, we successfully identified significant markers for CD8+ naive (BACH2), CD16+ (CD16), and CD14+ (VCAN) cell populations. CONCLUSION: MLme serves as a valuable resource for leveraging ML to facilitate insightful data analysis and enhance research outcomes, while alleviating concerns related to complex coding scripts. The source code and a detailed tutorial for MLme are available at https://github.com/FunctionalUrology/MLme.

Rad52, a highly conserved eukaryotic protein, plays a crucial role in DNA repair, especially in double-strand break repair. Recent findings reveal that its distinct structural features, including a characteristic β-sheet and β-hairpin motif, are shared with the lambda phage single-strand annealing proteins, Redβ, indicating a common superfamily. Our analysis of over 10,000 single-strand annealing proteins (SSAPs) across all kingdoms of life supports this hypothesis, confirming their possession of the characteristic motif despite variations in size and composition. We found that archaea, representing only 1% of the studied proteins, exhibit most of these variations. Through the examination of four representative archaeal SSAPs, we elucidate the structural relationship between eukaryotic and bacterial SSAPs, highlighting differences in β-sheet size and β-hairpin complexity. Furthermore, we identify an archaeal SSAP with a structure nearly identical to the human variant and screen over 100 million unannotated proteins for potential SSAP candidates. Our computational analysis complements existing sequence with structural evidence supporting the suggested orthology among five SSAP families across all kingdoms: Rad52, Redβ, RecT, Erf, and Sak3.Competing Interest StatementThe authors have declared no competing interest.All data used to generate this work are accessible via the link provided below. This collection includes both raw and generated data. The raw data consists of PDB files for all SSAP proteins and their FASTA sequences. The generated data encompass BLAST results (sequence data), structural alignments (TM scores), taxonomy information, and family memberships from InterPro. Please refer to the following link to access the data.Raw data: ./rawData/[PDBs, Fasta]Alignment results: ./alignments/https://sharing.biotec.tu-dresden.de/index.php/s/vNrJ3aHLSUN6kZE

A protein can be represented by its primary, secondary, or tertiary structure. With recent advances in AI, there is now as much tertiary as primary structural data available. Fast and accurate search methods exist for both types of data, with searches over both representations being highly precise. However, primary structure data can sometimes be incomplete. As a result, tertiary structure has become the gold standard for remote homology detection.How does secondary structure perform in remote homology detection? Secondary structure interprets proteins as a sequence using an alphabet representing helices, strands, or loops. It shares its sequential nature with primary structure while retaining topological information similar to tertiary structure.To assess the effectiveness of secondary structure in remote homology detection, we devised a challenging classification task aimed at determining the superfamily membership of very distantly related protein domains. We used benchmarks from the CATH and SCOP databases and evaluated sequence and structure alignment algorithms on primary, secondary, and tertiary structures.As expected, both basic and advanced sequence alignment algorithms applied to primary structure achieved high precision, but their overall area under the curve was lower compared to the gold standard of structural alignment using tertiary structure.Surprisingly, a simple string comparison algorithm applied to secondary structure performed close to the gold standard. This result supports the hypothesis that key structural information is already encoded in secondary structure and suggests that secondary structure may be a promising representation to use when high-confidence structural data is unavailable, such as in cases involving protein flexibility and disorder.Competing Interest StatementThe authors have declared no competing interest.

Coarse-Grained Reconfigurable Architectures (CGRAs) are one of the promising solutions to be employed in power-hungry edge devices owing to providing a good balance between reconfigurability, performance and energy-efficiency. Most of the proposed CGRAs feature a homogeneous set of processing elements (PEs) which all support the same set of operations. Homogeneous PEs can lead to high unwanted power consumption. As application benchmarks utilize different operations irregularly, heterogeneous PE design is a powerful approach to reduce power consumption of CGRA. In this paper, we propose DA-CGRA, a domain-aware CGRA tailored to signal processing applications. To extract heterogeneous architecture, first, a set of signal processing applications has been profiled to derive the requirements of the applications in terms of type of operations, number of operations and memory usage. Then, domain-specific PEs are designed using Verilog RTL based on the profiling results. We have selected spatio-temporal or spatial execution model based on the application features to increase the overall performance and efficiency. Experimental results demonstrate DA-CGRA outperforms FLEX and RipTide state-of-the-art CGRAs in terms of energy-efficiency by 23% and 38%, respectively. Moreover, DA-CGRA can achieve 3.2x performance improvement over HM-HvCUBE.

The BigCode project is an open-scientific collaboration working on the responsible development of large language models for code. This tech report describes the progress of the collaboration until December 2022, outlining the current state of the Personally Identifiable Information (PII) redaction pipeline, the experiments conducted to de-risk the model architecture, and the experiments investigating better preprocessing methods for the training data. We train 1.1B parameter models on the Java, JavaScript, and Python subsets of The Stack and evaluate them on the MultiPL-E text-to-code benchmark. We find that more aggressive filtering of near-duplicates can further boost performance and, surprisingly, that selecting files from repositories with 5+ GitHub stars deteriorates performance significantly. Our best model outperforms previous open-source multilingual code generation models (InCoder-6.7B and CodeGen-Multi-2.7B) in both left-to-right generation and infilling on the Java, JavaScript, and Python portions of MultiPL-E, despite being a substantially smaller model. All models are released under an OpenRAIL license at https://hf.co/bigcode.

The impact of the spring climate on the Northern Hemispheretextquoterights summer vegetation activity and extremes has been extensively researched, but less attention has been devoted to whether and how the winter climate may additionally influence vegetation extremes in the summer. Here, we provide insights into the influence of winter temperature and precipitation on summer vegetation extremes in the Northern Hemisphere. To do this, we identify positive and negative extremes in the summer leaf area index (LAI, a proxy for vegetation activity) and assess winter effects on those extremes using logistic regression at the regional scale. Over a quarter of the regions in the Northern Hemisphere show strong winter climate preconditioning on summer LAI extremes, which is typically stronger for croplands than forests. In regions with strong winter preconditioning, the spring LAI mediates the link between winter climate and summer LAI extremes through the ecological memory in seasonal legacy effects. Our findings suggest that extremely low summer LAI in both croplands and forests is preconditioned by colder and drier winters, while extremely high summer LAI in forests is associated with warmer and wetter winters. For low summer LAI in croplands, warmer winters are associated with an increased likelihood of extremes in mid-latitude regions and a reduced likelihood in high-latitude regions. Consideration of winter preconditioning effects may improve our understanding of inter-annual variability of vegetation activity and support agricultural and land management practitioners in anticipating the detrimental effects of winter on crop yields and forest conditions.

To capture the extremal behaviour of complex environmental phenomena in practice, flexible techniques for modelling tail behaviour are required. In this paper, we introduce a variety of such methods, which were used by the Lancopula Utopiversity team to tackle the EVA (2023) Conference Data Challenge. This data challenge was split into four challenges, labelled C1-C4. Challenges C1 and C2 comprise univariate problems, where the goal is to estimate extreme quantiles for a non-stationary time series exhibiting several complex features. For these, we propose a flexible modelling technique, based on generalised additive models, with diagnostics indicating generally good performance for the observed data. Challenges C3 and C4 concern multivariate problems where the focus is on estimating joint probabilities. For challenge C3, we propose an extension of available models in the multivariate literature and use this framework to estimate joint probabilities in the presence of non-stationary dependence. Finally, for challenge C4, which concerns a 50-dimensional random vector, we employ a clustering technique to achieve dimension reduction and use a conditional modelling approach to estimate extremal probabilities across independent groups of variables.

Inspired by the idea of RDF Redux, an RDF extension suggested by Pat Hayes, RDF Surfaces were recently developed and further specified by a W3C community group. The idea of RDF Surfaces is to add negation and explicit existential quantification to RDF and thereby gain the expressivity of first-order logic. RDF Surfaces come with a syntax and even with first implementations, but the semantics has so far only been defined informally. In this paper we aim to close this gap: we map RDF Surface graphs to first-order logic formulae and thereby define their semantics. We show that, restricted to RDF graphs, this semantics preserves simple entailment. That is, each RDF graph which entails another in its first-order translation, also entails this graph according to RDF's simple entailment and vice versa. To test whether this semantics fully meets the informal specification, we furthermore provide rs2fol, an implementation which follows our mapping and translates RDF Surfaces in N3-based syntax to first-order logic in TPTP syntax. We apply this implementation on the various examples collected on the Web page of the RDF Surfaces reasoner EYE, run them with the theorem prover Vampire and compare the results with those of EYE. With the exception of a different understanding of lists -- EYE treats these as first-class citizens -- results of both approaches coincide. We thus provides a tool for entailment checking which is conform to the current specification. This tool will help future developers of RDF Surfaces reasoners to test their derivations for correctness and the community as a whole to better understand the logic and -- if needed -- to refine or even restrict it.

The growing use of Machine Learning and Artificial Intelligence (AI), particularly Large Language Models (LLMs) like OpenAI's GPT series, leads to disruptive changes across organizations. At the same time, there is a growing concern about how organizations handle personal data. Thus, privacy policies are essential for transparency in data processing practices, enabling users to assess privacy risks. However, these policies are often long and complex. This might lead to user confusion and consent fatigue, where users accept data practices against their interests, and abusive or unfair practices might go unnoticed. LLMss can be used to assess privacy policies for users automatically. In this interdisciplinary work, we explore the challenges of this approach in three pillars, namely technical feasibility, ethical implications, and legal compatibility of using LLMs to assess privacy policies. Our findings aim to identify potential for future research, and to foster a discussion on the use of LLM technologies for enabling users to fulfil their important role as decision-makers in a constantly developing AI-driven digital economy.

Concrete domains have been introduced in description logic (DL) to enable reference to concrete objects (such as numbers) and predefined predicates on these objects (such as numerical comparisons) when defining concepts. The primary research goal in this context was to find restrictions on the concrete domain such that its integration into certain DLs preserves decidability or tractability. In this article, we complement these investigations by studying the abstract expressive power of both first-order and description logics extended with concrete domains, i.e., we analyze which classes of first-order interpretations can be expressed using these logics, compared to what their counterparts without concrete domains can express. We demonstrate that, under natural conditions on the concrete domain D (which also play a role for decidability), extensions of first-order logic (FOL) or the well-known DL ALC with D share important formal characteristics with FOL, such as the compactness and the Löwenheim-Skolem properties. Nevertheless, these conditions do not ensure that the abstract expressive power of the extensions we consider is contained in that of FOL, though in some cases it is. To be more precise, we show, on the one hand, that unary concrete domains leave the abstract expressive power within FOL if we are allowed to introduce auxiliary predicates. As a by-product, we obtain (semi-)decidability results for some fragments of FOL extended with the concrete domains considered in this article. On the other hand, we show that the ability to express equality between elements of D, another condition employed in the context of showing decidability of ALC(D), is sufficient to push the abstract expressive power of most first-order fragments with concrete domains beyond that of FOL. While for such concrete domains D decidability is retained for ALC(D), we show that the availability of equality in D causes undecidability of the two-variable fragment of FOL(D), although the two-variable fragment of FOL is decidable.

Concrete domains have been introduced in description logic (DL) to enable reference to concrete objects (such as numbers) and predefined predicates on these objects (such as numerical comparisons) when defining concepts. The primary research goal in this context was to find restrictions on the concrete domain such that its integration into certain DLs preserves decidability or tractability. In this article, we complement these investigations by studying the abstract expressive power of both first-order and description logics extended with concrete domains, i.e., we analyze which classes of first-order interpretations can be expressed using these logics, compared to what their counterparts without concrete domains can express. We demonstrate that, under natural conditions on the concrete domain D (which also play a role for decidability), extensions of first-order logic (FOL) or the well-known DL ALC with D share important formal characteristics with FOL, such as the compactness and the Löwenheim-Skolem properties. Nevertheless, these conditions do not ensure that the abstract expressive power of the extensions we consider is contained in that of FOL, though in some cases it is. To be more precise, we show, on the one hand, that unary concrete domains leave the abstract expressive power within FOL if we are allowed to introduce auxiliary predicates. As a by-product, we obtain (semi-)decidability results for some fragments of FOL extended with the concrete domains considered in this article. On the other hand, we show that the ability to express equality between elements of D, another condition employed in the context of showing decidability of ALC(D), is sufficient to push the abstract expressive power of most first-order fragments with concrete domains beyond that of FOL. While for such concrete domains D decidability is retained for ALC(D), we show that the availability of equality in D causes undecidability of the two-variable fragment of FOL(D), although the two-variable fragment of FOL is decidable.

Concrete domains have been introduced in description logic (DL) to enable reference to concrete objects (such as numbers) and predefined predicates on these objects (such as numerical comparisons) when defining concepts. The primary research goal in this context was to find restrictions on the concrete domain such that its integration into certain DLs preserves decidability or tractability. In this paper, we investigate the abstract expressive power of both first-order and description logics extended with concrete domains, i.e., we analyze which classes of first-order interpretations can be expressed using these logics, compared to what first-order logic without concrete domains can express. We demonstrate that, under natural conditions on the concrete domain D (which also play a role for decidability), extensions of first-order logic (FOL) or the well-known DL ALC with D share important formal characteristics with FOL, such as the compactness and the Löwenheim-Skolem properties. Nevertheless, their abstract expressive power need not be contained in that of FOL, though in some cases it is. To be more precise, we show, on the one hand, that unary concrete domains leave the abstract expressive power within FOL if we are allowed to introduce auxiliary predicates. On the other hand, we exhibit a class of concrete domains that push the abstract expressive power beyond that of FOL. As a by-product of these investigations, we obtain (semi-)decidability results for some of the logics with concrete domains considered in this paper.

Unification has been introduced in Description Logic (DL) as a means to detect redundancies in ontologies. In particular, it was shown that testing unifiability in the DL $$backslashmathcalEbackslashmathcalL$$ELis an NP-complete problem, and this result has been extended in several directions. Surprisingly, it turned out that the complexity increases to PSpace if one disallows the use of the top concept in concept descriptions. Motivated by features of the medical ontology SNOMED CT, we extend this result to a setting where the top concept is disallowed, but there is a background ontology consisting of restricted forms of concept and role inclusion axioms. We are able to show that the presence of such axioms does not increase the complexity of unification without top, i.e., testing for unifiability remains a PSpace-complete problem.

In the realm of artificial intelligence (AI) and machine learning (ML), the scarcity of robust and diverse datasets often poses a significant challenge, prompting the need for effective data generation methods. This paper presents an evaluation of tabular data generation techniques on the DaFne platform, centered around a predictive maintenance case study for bridges. The DaFne platform offers a variety of tabular data generation functionalities, including rule-based creation, data fusion (with weather data), and data reproduction. We investigate the utility of these functionalities across different machine learning models for the prediction of bridge conditions. Our analysis includes a descriptive statistical comparison of real and synthetic data. Additionally, we explore the utility of original, weather, and synthetic datasets. We do this through the lens of ML models like MLR, XGBoost, CNN, and GRU, performing a predictive maintenance algorithm on these datasets. Our results indicate that while the inclusion of weather data did not significantly enhance predictive performance, the synthetic dataset shows satisfactory quality. However, the synthetic data's performance is lower than the original data in predictive maintenance tasks, with differences observed in models heavily reliant on sequential data. This research underscores the potential of the DaFne platform in generating high-quality synthetic data. It also highlights areas for future improvement and offers valuable insights for advancing data generation and analysis techniques in predictive maintenance and other AI applications.

The decline of segetal flora species across Europe, driven by intensified agricultural practices, is impacting other taxa and ecosystem functions. Result-based payments to farmers offer an effective solution to conserve these species, but the high cost of biodiversity monitoring remains a challenge. In this study, we conducted UAV flights with an RGB camera and used the deep learning model YOLO to detect these species in four winter barley fields under different management intensities in Germany. Field measurements of plant traits were used to evaluate their impact on species detectability. Additionally, we investigated the potential of spatial co-occurrence and canopy height heterogeneity to predict the presence of species difficult to detect by UAVs. We found that half of the species observed could be remotely detected, with a minimum ground sampling distance (GSD) of 1.22 mm required for accurate annotation. The same detection ratio was estimated for key indicator species not present in our study area based on trait information. Plant height was crucial for species detection, with accuracy ranging between 49–100 %. YOLO models effectively predicted species from images taken at 40 m, reducing the monitoring time to eight minutes per hectare. Co-occurrence with UAV-detectable species and canopy height heterogeneity proved promising for identifying areas where undetectable species are likely to occur, although further research is needed for landscape-level applications. Our study highlights the potential for large-scale, cost-effective monitoring of segetal flora species in agricultural landscapes, and provides valuable insights for developing robust textquoteleftsmart indicatorstextquoteright for future biodiversity monitoring.

In the future, roads will host a complex mix of automated and manually operated vehicles, along with vulnerable road users. However, most automotive user interfaces and human factors research focus on single-agent studies, where one human interacts with one vehicle. Only a few studies incorporate multi-agent setups. This workshop aims to (1) examine the current state of multi-agent research in the automotive domain, (2) serve as a platform for discussion toward more realistic multi-agent setups, and (3) discuss methods and practices to conduct such multi-agent research. The goal is to synthesize the insights from the AutoUI community, creating the foundation for advancing multi-agent traffic interaction research.

Large-scale power outages, such as those caused by extreme weather events, have a big impact on human behavior. A short power outage is merely a nuisance for most, and may not change people’s locations. An outage that lasts for a few hours can result in spoiled food and medical supplies, and people will have to restock spoiled items. Long outages result in temperatures outside tolerable levels in homes, and may prompt people to acquire supplies, such as generators and gas, or change location. The long outages during Winter Storm Uri in Texas resulted in millions of dollars in property damage due to freezing pipes. This level of damage is expected to result in a sharp increase in supply runs and contractor activity. In this paper, we present a tool to explore differences in visiting patterns before, during, and after power outages. It allows to compare different points of interest like medical facilities, grocery stores, hardware stores, and other types of businesses.

YouTube videos are a popular medium for online product reviews. They are not only informative and entertaining, but may also be perceived as quite credible under the viewer’s impression of a personal product demonstration by an expert. As the world’s largest online video platform, YouTube’s content is included prominently in the results of most general-purpose web search engines. Consequently, online marketeers are using classic Search Engine Optimization (SEO) techniques also for placing their video content in search engines. Over the years, we have noticed an ever increasing noise floor of low-quality SEO content in product search results and in this study, we show that this trend has spilled over into videos as well. We examine YouTube video reviews for several thousand products retrieved from three commercial search engines and conduct spam detection experiments based directly on the videos’ subtitle transcripts rather than relying on metadata and comments. We find that at least a third of the retrieved videos can be regarded as spam or low-quality productions. We are further able to distinguish these spam product reviews accurately from higher-quality videos with a semi-supervised n-gram classification approach.

Valued constraint satisfaction problems (VCSPs) constitute a large class of computational optimisation problems. It was shown recently that, over finite domains, every VCSP is in P or NP-complete, depending on the admitted cost functions. In this article, we study cost functions over countably infinite domains whose automorphisms form an oligomorphic permutation group. Our results include a hardness condition based on a generalisation of pp-constructability as known from classical CSPs and a polynomial-time tractability condition based on the concept of fractional polymorphisms. We then observe that the resilience problem for unions of conjunctive queries (UCQs) studied in database theory, under bag semantics, may be viewed as a special case of the VCSPs that we consider. We obtain a complexity dichotomy for the case of incidence-acyclic UCQs and exemplarily use our methods to determine the complexity of a query that had remained open in the literature. Further, we conjecture that our hardness and tractability conditions match for resilience problems for UCQs.

Teachers are a core element of the educational world and improving the teaching skills of future and current teachers has significant individual and societal impacts. This contribution reviews the educational environment in Cambodia which was the starting point of all activities in CONTESSA with reference to the Cambodian context. Based on document analyses, online-surveys, semi-structured interviews and focus groups with pre- and in-service teachers, teacher trainers and government representatives, this contribution discusses the challenges and opportunities of providing access to inclusive, equitable and contemporary teacher education to a wide range of pedagogues in Cambodia through the context of CONTESSA – Contemporary Teaching Skills for South Asia.

Concrete domains have been introduced in the context of Description Logics to allow references to qualitative and quantitative values. In particular, the class of ω-admissible concrete domains, which includes Allen's interval algebra, the region connection calculus (RCC8), and the rational numbers with ordering and equality, has been shown to yield extensions of ALC for which concept satisfiability w.r.t. a general TBox is decidable. In this paper, we present an algorithm based on type elimination and use it to show that deciding the consistency of an ALC(D) ontology is ExpTime-complete if the concrete domain D is ω-admissible and its constraint satisfaction problem is decidable in exponential time. While this allows us to reason with concept and role assertions, we also investigate feature assertions f(a,c) that can specify a constant c as the value of a feature f for an individual a. We show that, under conditions satisfied by all known ω-admissible domains, we can add feature assertions without affecting the complexity

In this paper, we present a research platform to support studying collaboration in hybrid and co-located scenarios. Mixed-presence collaboration includes various novel and exciting use cases, such as situated and immersive data analysis by multiple users. However, research in this emerging field is hindered by the technical complexity of the setups and often requires re-implementation of common features. We address this issue by contributing a toolkit and research platform for mixed-presence collaboration that serves as an extensible baseline implementation and enables fast prototyping for user studies in collaborative mixed reality. Furthermore, our platform provides adjustable parameters, such as types of avatars, audio source placement, or the amount of simulated network latency. This way, developers are supported in making design choices regarding typical, re-occurring technical challenges.

In formal argumentation one aims for intuitive and concise justifications for the acceptance of arguments. Discussion games and dispute trees are established methods to obtain such a justification. However, so far these techniques are based on instantiating the knowledge base into graph-based Dung style abstract argumentation frameworks (AFs). These instantiations are known to produce frameworks with a large number of arguments and thus also yield long discussion games and large dispute trees. To obtain more concise justifications for argument acceptance, we propose to instantiate the knowledge base as an argumentation framework with collective attacks (SETAF). Remarkably, this approach yields smaller frameworks compared to traditional AF instantiation, while exhibiting increased expressive power. We then introduce discussion games and dispute trees tailored to SETAFs, show that they correspond to credulous acceptance w.r.t. the well-known preferred semantics, analyze and tune them w.r.t. the size, and compare the two notions. Finally, we illustrate how our findings apply to assumption-based argumentation.

In recent years, artificial intelligence (AI) has deeply impacted various fields, including Earth system sciences. Here, AI improved weather forecasting, model emulation, parameter estimation, and the prediction of extreme events. However, the latter comes with specific challenges, such as developing accurate predictors from noisy, heterogeneous and limited annotated data. This paper reviews how AI is being used to analyze extreme events (like floods, droughts, wildfires and heatwaves), highlighting the importance of creating accurate, transparent, and reliable AI models. We discuss the hurdles of dealing with limited data, integrating information in real-time, deploying models, and making them understandable, all crucial for gaining the trust of stakeholders and meeting regulatory needs. We provide an overview of how AI can help identify and explain extreme events more effectively, improving disaster response and communication. We emphasize the need for collaboration across different fields to create AI solutions that are practical, understandable, and trustworthy for analyzing and predicting extreme events. Such collaborative efforts aim to enhance disaster readiness and disaster risk reduction.

Managing water resources in regions with high climate variability and frequent extreme weather events poses challenges for policymakers. To facilitate water allocation in these cases, participatory and collaborative decision-making approaches have become common. However, the evaluation of these approaches is hindered by the lack of structured methods and data to understand them. To address this knowledge gap, we propose a novel methodology that leverages text data to identify key topics, conflicts, and influential actors that shape water allocation dynamics. Our methodology is tested using records of 1020 water basin committee meetings held between 1997 and 2021 across 12 basin committees in Ceará, Brazil—a region known for its extensive history of droughts that have impacted water governance. To uncover key water management issues discussed during these meetings, we employed a three-step …

The increased frequency and magnitude of natural hazards might significantly increase social, economic, and health impacts on society in the next decades. Existing studies and databases of natural hazard impacts have several limitations, such as (1) a low level of detail on how people were affected;(2) an underestimation of the impacts;(3) a limited geographical range; and (4) a lack of information on the source of the data. However, scientific publications, reports, and handbooks compose a large data repository that can provide valuable and trustworthy information on natural hazards. We are building a global database on the impacts of natural hazards that have been documented since 1950 in the scientific literature. We mapped global research on climatological, hydrological, and meteorological extremes, such as heatwaves and floods. We retrieved over 40 thousand full-text open-access papers from …

The increased frequency and magnitude of natural hazards might significantly increase social, economic, and health impacts on society in the next decades. Existing studies and databases of natural hazard impacts have several limitations, such as (1) a low level of detail on how people were affected;(2) an underestimation of the impacts;(3) a limited geographical range; and (4) a lack of information on the source of the data. However, scientific publications, reports, and handbooks compose a large data repository that can provide valuable and trustworthy information on natural hazards. We are building a global database on the impacts of natural hazards that have been documented since 1950 in the scientific literature. We mapped global research on climatological, hydrological, and meteorological extremes, such as heatwaves and floods. We retrieved over 40 thousand full-text open-access papers from …

A cascade of drought impacts refers to a series of interconnected events that trigger a chain reaction of impacts, extending beyond water scarcity, to affect agricultural production, socio-economic factors, and the environment. This paper aims to understand the role of society in mitigating drought impacts, particularly through policy responses. Conducting a case study in Ceará state, northeast Brazil, we used a global rare dataset of continuously drought monitoring, complemented by interviews with smallholder farmers and agricultural extension technicians. Additionally, we analyzed policy documents related to public policies implemented at the local level. Employing a classification of drought impacts as our analytical framework, our findings indicate that socio-environmental-economic impacts of drought are less frequently reported, suggesting that development policies are mitigating cascading effects on livelihoods. Most impacts are associated with hydrological impacts of drought, suggesting unintended consequences of investments in increasing water supply. We emphasize the significant contribution of public policies to mitigating the cascading effects of drought, which do not necessarily involve increasing water availability, but strengthen the local economy.

Recent studies have explored the delay propagation network among airports by inferring Granger causality among time series using linear methods. Granger causality indicates whether delays at one airport can help predict delays at itself or other airports and thus help to understand the interconnection among airports. This paper makes inferences about the Granger causality relation among airports by applying a nonlinear multilayer perceptron method to the arrival delay time series in Europe and China. We find that the propagation results can be overestimated with data input containing early arrival flights, besides delayed flights. Europe presents a smaller magnitude of delay propagation than China, considering only delayed flights, indicated by the link densities of the networks with the same penalty parameter. Unlike some recent research findings, our results suggest that large airports have more out-degree and more impact in the delay propagation network. These results can help predict and understand delay propagation in daily operations or simulation environments.

Tree mortality has escalated worldwide in recent years due to climate warming and unprecedented drought events. However, mapping tree mortality across forest ecosystems has not yet been achieved. Aerial photos provide opportunities to reveal the spatial and spectral characteristics of canopy death at local to landscape scales. In this work, we present a deep learning model for mapping tree mortality from aerial photos in various forested ecosystems across Europe. This model builds on a baseline model trained with data on dead tree canopies from California using sub-meter resolution aerial photos and allows the use of various spatial resolutions of the input images (ranging from 10 to 60 cm). By comparing our results to ground observations and/or state-of-the-art forest disturbance and loss products, we will discuss the advantages and limitations of aerial photo-based tree mortality mapping. The proposed …

The Interplanetary File System (IPFS) [1] is a popular Peer-to-Peer (P2P) overlay network with a focus on content-addressed data exchange. While IPFS offers decentralized file storage, improving resilience, it suffers from privacy challenges [3]. In particular Bitswap, the data exchange protocol suffers under these challenges. That is, Bitswap contacts all neighbors for content discovery revealing interest to many participants.

Understanding why some individuals age faster than others is essential to evolutionary biology and geroscience, but measuring variation in biological age is difficult. One solution may lie in measuring gut microbiome composition because microbiota change with many age-related factors (e.g., immunity and behavior). Here we create a microbiome-based age predictor using 13,563 gut microbial profiles from 479 wild baboons collected over 14 years. The resulting textquotedblleftmicrobiome clocktextquotedblright predicts host chronological age. Deviations from the clocktextquoterights predictions are linked to demographic and socio-environmental factors that predict baboon health and survival: animals who appear old-for-age tend to be male, sampled in the dry season (for females), and high social status (both sexes). However, an individualtextquoterights textquotedblleftmicrobiome agetextquotedblright does not predict the attainment of developmental milestones or lifespan. Hence, the microbiome clock accurately reflects age and some social and environmental conditions, but not the pace of development or mortality risk.Competing Interest StatementThe authors have declared no competing interest.

The accuracy of neural networks has greatly improved across various domains over the past years. Their ever-increasing complexity, however, leads to prohibitively high energy demands and latency in von Neumann systems. Several computing-in-memory (CIM) systems have recently been proposed to overcome this, but trade-offs involving accuracy, hardware reliability, and scalability for large models remain a challenge. Additionally, for some CIM designs, the activation movement still requires considerable time and energy. This paper explores the combination of algorithmic optimizations for ternary weight neural networks and associative processors (APs) implemented using racetrack memory (RTM). We propose a novel compilation flow to optimize convolutions on APs by reducing their arithmetic intensity. By leveraging the benefits of RTM-based APs, this approach substantially reduces data transfers within the memory while addressing accuracy, energy efficiency, and reliability concerns. Concretely, our solution improves the energy efficiency of ResNet-18 inference on ImageNet by 7.5x compared to crossbar in-memory accelerators while retaining software accuracy.

Virological plaque assays are the primary method for quantifying infectious particles in a suspension, achieved by incubating a serial dilution of the virus with a monolayer of indicator cells. Existing software tools for quantification of plaque assay images lack modularity, show measurements disagreements or are closed-source - a common hurdle in BioImage analysis. We introduce PyPlaque, an open-source Python package focusing on flexibility and modularity rather than a bulky graphic user interface. Unlike previous methods, an abstracted architecture using object-oriented programming allows accommodation of various experimental containers and specimen carriers as data structures while focusing on phenotype-specific information. Aligned with the logical flow of experimental design and desired quantifications, it delivers insights at multiple granularity levels, facilitating detailed analysis. We demonstrate how this approach allows to focus on alleviating the disagreement in measurements. Furthermore, similar design is generalisable to diverse datasets in various biological contexts that fit our structural paradigm.Competing Interest StatementThe authors have declared no competing interest.

Foliar traits such as specific leaf area (SLA), leaf nitrogen (N), and phosphorus (P) concentrations play important roles in plant economic strategies and ecosystem functioning. Various global maps of these foliar traits have been generated using statistical upscaling approaches based on in-situ trait observations. Here, we intercompare such global upscaled foliar trait maps at 0.5° spatial resolution (six maps for SLA, five for N, three for P), categorize the upscaling approaches used to generate them, and evaluate the maps with trait estimates from a global database of vegetation plots (sPlotOpen). We disentangled the contributions from different plant functional types (PFTs) to the upscaled maps and quantified the impacts of using different plot-level trait metrics on the evaluation with sPlotOpen: community weighted mean (CWM) and top-of-canopy weighted mean (TWM). We found that the global foliar trait maps of SLA and N differ drastically and fall into two groups that are almost uncorrelated (for P only maps from one group were available). The primary factor explaining the differences between these groups is the use of PFT information combined with remote sensing-derived land cover products in one group while the other group mostly relied on environmental predictors alone. The maps that used PFT and corresponding land cover information exhibit considerable similarities in spatial patterns that are strongly driven by land cover. The maps not using PFTs show a lower level of similarity and tend to be strongly driven by individual environmental variables. Upscaled maps of both groups were moderately correlated to sPlotOpen data aggregated to the grid-cell level (R = 0.2–0.6) when processing sPlotOpen in a way that is consistent with the respective trait upscaling approaches, including the plot-level trait metric (CWM or TWM) and the scaling to the grid cells with or without accounting for fractional land cover. The impact of using TWM or CWM was relevant, but considerably smaller than that of the PFT and land cover information. The maps using PFT and land cover information better reproduce the between-PFT trait differences of sPlotOpen data, while the two groups performed similarly in capturing within-PFT trait variation. Our findings highlight the importance of explicitly accounting for within-grid-cell trait variation, which has important implications for applications using existing maps and future upscaling efforts. Remote sensing information has great potential to reduce uncertainties related to scaling from in-situ observations to grid cells and the regression-based mapping steps involved in the upscaling.

The recently proposed MA-BBOB function generator provides a way to create numerical black-box benchmark problems based on the well-established BBOB suite. Initial studies on this generator highlighted its ability to smoothly transition between the component functions, both from a low-level landscape feature perspective, as well as with regard to algorithm performance. This suggests that MA-BBOB-generated functions can be an ideal testbed for automated machine learning methods, such as automated algorithm selection (AAS). In this paper, we generate 11800 functions in dimensions $d=2$ and $d=5$, respectively, and analyze the potential gains from AAS by studying performance complementarity within a set of eight algorithms. We combine this performance data with exploratory landscape features to create an AAS pipeline that we use to investigate how to efficiently select training sets within this space. We show that simply using the BBOB component functions for training yields poor test performance, while the ranking between uniformly chosen and diversity-based training sets strongly depends on the distribution of the test set.

Assumption-based argumentation (ABA) is a powerful defeasible reasoning formalism which is based on the interplay of assumptions, their contraries, and inference rules. ABA with preferences (ABA+) generalizes the basic model by allowing qualitative comparison between assumptions. The integration of preferences however comes with a cost. In ABA+, the evaluation under two central and well-established semantics---grounded and complete semantics---is not guaranteed to yield an outcome. Moreover, while ABA frameworks without preferences allow for a graph-based representation in Dung-style frameworks, an according instantiation for general ABA+ frameworks has not been established so far. In this work, we tackle both issues: First, we develop a novel abstract argumentation formalism based on set-to-set attacks. We show that our so-called Hyper Argumentation Frameworks (HYPAFs) capture ABA+. Second, we propose relaxed variants of complete and grounded semantics for HYPAFs that yield an extension for all frameworks by design, while still faithfully generalizing the established semantics of Dung-style Argumentation Frameworks. We exploit the newly established correspondence between ABA+ and HYPAFs to obtain variants for grounded and complete ABA+ semantics that are guaranteed to yield an outcome. Finally, we discuss basic properties and provide a complexity analysis. Along the way, we settle the computational complexity of several ABA+ semantics.

Argumentation frameworks (AFs) are a key formalism in AI research. Their semantics have been investigated in terms of principles, which define characteristic properties in order to deliver guidance for analyzing established and developing new semantics. Because of the simple structure of AFs, many desired properties hold almost trivially, at the same time hiding interesting concepts behind syntactic notions. We extend the principle-based approach to argumentation frameworks with collective attacks (SETAFs) and provide a comprehensive overview of common principles for their semantics. Our analysis shows that investigating principles based on decomposing the given SETAF (e.g. directionality or SCC-recursiveness) poses additional challenges in comparison to usual AFs. We introduce the notion of the reduct as well as the modularization principle for SETAFs which will prove beneficial for this kind of investigation. We then demonstrate how our findings can be utilized for incremental computation of extensions and show how we can use graph properties of the frameworks to speed up these algorithms.

BACKGROUND: The overall survival of patients with advanced and refractory oesophageal squamous cell carcinoma, mostly aged 65 years and older, is poor. Treatment with PD-1 antibodies showed improved progression-free survival and overall survival. We assessed the safety and efficacy of combined nivolumab and ipilimumab therapy in this population. METHODS: This multicentre, open-label, phase 2 trial done in 32 sites in Germany included patients aged 65 years and older with oesophageal squamous cell carcinoma and disease progression or recurrence following first-line therapy. Patients were treated with nivolumab (240 mg fixed dose once every 2 weeks, intravenously) in the safety run-in phase and continued with nivolumab and ipilimumab (nivolumab 240 mg fixed dose once every 2 weeks and ipilimumab 1 mg/kg once every 6 weeks, intravenously). The primary endpoint was overall survival, which was compared with a historical cohort receiving standard chemotherapy in the intention-to-treat population. This study is registered with ClinicalTrials.gov, NCT03416244. FINDINGS: Between March 2, 2018, and Aug 20, 2020, we screened 75 patients with advanced oesophageal squamous cell carcinoma. We enrolled 66 patients (50 [76%] men and 16 [24%] women; median age 70·5 years [IQR 67·0-76·0]), 44 (67%) of whom received combined nivolumab and ipilimumab therapy and 22 (33%) received nivolumab alone. Median overall survival time at the prespecified data cutoff was 7·2 months (95% CI 5·7-12·4) and significantly higher than in a historical cohort receiving standard chemotherapy (p=0·0063). The most common treatment-related adverse events were fatigue (12 [29%] of 42), nausea (11 [26%]), and diarrhoea (ten [24%]). Grade 3-5 treatment-related adverse events occurred in 13 (20%) of 66 patients. Treatment-related death occurred in one patient with bronchiolitis obliterans while on nivolumab and ipilimumab treatment. INTERPRETATION: Patients aged at least 65 years, with advanced oesophageal squamous cell carcinoma might benefit from combined nivolumab and ipilimumab therapy in second-line treatment. FUNDING: Bristol Myers Squibb.

Clinical research relies on high-quality patient data, however, obtaining big data sets is costly and access to existing data is often hindered by privacy and regulatory concerns. Synthetic data generation holds the promise of effectively bypassing these boundaries allowing for simplified data accessibility and the prospect of synthetic control cohorts. We employed two different methodologies of generative artificial intelligence - CTAB-GAN+ and normalizing flows (NFlow) - to synthesize patient data derived from 1606 patients with acute myeloid leukemia, a heterogeneous hematological malignancy, that were treated within four multicenter clinical trials. Both generative models accurately captured distributions of demographic, laboratory, molecular and cytogenetic variables, as well as patient outcomes yielding high performance scores regarding fidelity and usability of both synthetic cohorts (n = 1606 each). Survival analysis demonstrated close resemblance of survival curves between original and synthetic cohorts. Inter-variable relationships were preserved in univariable outcome analysis enabling explorative analysis in our synthetic data. Additionally, training sample privacy is safeguarded mitigating possible patient re-identification, which we quantified using Hamming distances. We provide not only a proof-of-concept for synthetic data generation in multimodal clinical data for rare diseases, but also full public access to synthetic data sets to foster further research.

The fetal membranes are essential for the maintenance of pregnancy, and their integrity until parturition is critical for both fetal and maternal health. Preterm premature rupture of the membranes (pPROM) is known to be an indicator of preterm birth, but the underlying architectural and mechanical changes that lead to fetal membrane failure are not yet fully understood. The aim of this study was to gain new insights into the anatomy of the fetal membrane and to establish a tissue processing and staining protocol suitable for future prospective cohort studies.In this proof of principle study, we collected fetal membranes from women undergoing vaginal delivery or cesarean section. Small membrane sections were then fixed, stained for nucleic acids, actin, and collagen using fluorescent probes, and subsequently imaged in three dimensions using a spinning disk confocal microscope.Four fetal …

PURPOSE: The fetal membranes are essential for the maintenance of pregnancy, and their integrity until parturition is critical for both fetal and maternal health. Preterm premature rupture of the membranes (pPROM) is known to be an indicator of preterm birth, but the underlying architectural and mechanical changes that lead to fetal membrane failure are not yet fully understood. The aim of this study was to gain new insights into the anatomy of the fetal membrane and to establish a tissue processing and staining protocol suitable for future prospective cohort studies. METHODS: In this proof of principle study, we collected fetal membranes from women undergoing vaginal delivery or cesarean section. Small membrane sections were then fixed, stained for nucleic acids, actin, and collagen using fluorescent probes, and subsequently imaged in three dimensions using a spinning disk confocal microscope. RESULTS: Four fetal membranes of different types were successfully processed and imaged after establishing a suitable protocol. Cellular and nuclear outlines are clearly visible in all cases, especially in the uppermost membrane layer. Focal membrane (micro) fractures could be identified in several samples. CONCLUSION: The presented method proves to be well suited to determine whether and how the occurrence of membrane (micro) fractures and cellular jamming correlate with the timing of membrane rupture and the mode of delivery. In future measurements, this method could be combined with mechanical probing techniques to compare optical and mechanical sample information.

Abstract Soil is central to the complex interplay among biodiversity, climate, and society. This paper examines the interconnectedness of soil biodiversity, climate change, and societal impacts, emphasizing the urgent need for integrated solutions. Human-induced biodiversity loss and climate change intensify environmental degradation, threatening human well-being. Soils, rich in biodiversity and vital for ecosystem function regulation, are highly vulnerable to these pressures, affecting nutrient cycling, soil fertility, and resilience. Soil also crucially regulates climate, influencing energy, water cycles, and carbon storage. Yet, climate change poses significant challenges to soil health and carbon dynamics, amplifying global warming. Integrated approaches are essential, including sustainable land management, policy interventions, technological innovations, and societal engagement. Practices like agroforestry and organic farming improve soil health and mitigate climate impacts. Effective policies and governance are crucial for promoting sustainable practices and soil conservation. Recent technologies aid in monitoring soil biodiversity and implementing sustainable land management. Societal engagement, through education and collective action, is vital for environmental stewardship. By prioritizing interdisciplinary research and addressing key frontiers, scientists can advance understanding of the soil biodiversity–climate change–society nexus, informing strategies for environmental sustainability and social equity.

The first International Workshop on Open Web Search (WOWS) was held on Thursday, March 28th, at ECIR 2024 in Glasgow, UK. The full-day workshop had two calls for contributions: the first call aimed at scientific contributions to building, operating, and evaluating search engines cooperatively and the cooperative use of the web as a resource for researchers and innovators. The second call for implementations of retrieval components aimed to gain practical experience with joint, cooperative evaluation of search engines and their components. In total, 2 papers were accepted for the first call, and 11 software components were submitted for the second. The workshop ended with breakout sessions on how the Open-WebSearch.eu project can incorporate collaborative evaluations and a hub of search engines.Date: 28 March 2024.Website: https://opensearchfoundation.org/wows2024/.

Automated biomechanical testing has great potential for the development of VR applications, as initial insights into user behaviour can be gained in silico early in the design process. In particular, it allows prediction of user movements and ergonomic variables, such as fatigue, prior to conducting user studies. However, there is a fundamental disconnect between simulators hosting state-of-the-art biomechanical user models and simulators used to develop and run VR applications. Existing user simulators often struggle to capture the intricacies of real-world VR applications, reducing ecological validity of user predictions. In this paper, we introduce sim2vr, a system that aligns user simulation with a given VR application by establishing a continuous closed loop between the two processes. This, for the first time, enables training simulated users directly in the same VR application that real users interact with. We demonstrate that sim2vr can predict differences in user performance, ergonomics and strategies in a fast-paced, dynamic arcade game. In order to expand the scope of automated biomechanical testing beyond simple visuomotor tasks, advances in cognitive models and reward function design will be needed.

B-cell maturation antigen (BCMA)-targeting chimeric antigen receptor (CAR) T cells revolutionized the treatment of relapsed/refractory multiple myeloma (RRMM). However, data on cellular (CAR) T cell dynamics and the association with response, resistance or the occurrence of cytokine release syndrome (CRS) are limited. Therefore, we performed a comprehensive flow cytometry analysis of 27 RRMM patients treated with Idecabtagene vicleucel (Ide-cel) to assess the expansion capacity, persistence and effects on bystander cells of BCMA-targeting CAR T cells. Additionally, we addressed side effects, like cytokine release syndrome (CRS) and cytopenia. Our results show that in vivo expansion of CD8+ CAR T cells is correlated to response, however persistence is not essential for durable remission in RRMM patients. In addition, our data provide evidence, that an increased fraction of CD8+ T cells at day of leukapheresis in combination with successful lymphodepletion positively influence the outcome. We show that patients at risk for higher-grade CRS can be identified already prior to lymphodepletion. Our extensive characterization contributes to a better understanding of the dynamics and effects of BCMA-targeting CAR T cells, in order to predict the response of individual patients as well as side effects, which can be counteracted at an early stage or even prevented.

Privacy policies are expected to inform data subjects about their data protection rights and should explain the data controller's data management practices. Privacy policies only fulfill their purpose, if they are correctly interpreted, understood, and trusted by the data subject. This implies that a privacy policy is written in a fair way, e.g., it does not use polarizing terms, does not require a certain education, or does not assume a particular social background. We outline our approach to assessing fairness in privacy policies. We identify from fundamental legal sources and fairness research, how the dimensions informational fairness, representational fairness and ethics / morality are related to privacy policies. We propose options to automatically assess policies in these fairness dimensions, based on text statistics, linguistic methods and artificial intelligence. We conduct initial experiments with German privacy policies to provide evidence that our approach is applicable. Our experiments indicate that there are issues in all three dimensions of fairness. This is important, as future privacy policies may be used in a corpus for legal artificial intelligence models.

Natural Language Processing (NLP) plays an important role in our daily lives, particularly due to the enormous progress of Large Language Models (LLM). However, NLP has many fairness-critical use cases, e.g., as an expert system in recruitment or as an LLM-based tutor in education. Since NLP is based on human language, potentially harmful biases can diffuse into NLP systems and produce unfair results, discriminate against minorities or generate legal issues. Hence, it is important to develop a fairness certification for NLP approaches. We follow a qualitative research approach towards a fairness certification for NLP. In particular, we have reviewed a large body of literature on algorithmic fairness, and we have conducted semi-structured expert interviews with a wide range of experts from that area. We have systematically devised six fairness criteria for NLP, which can be further refined into 18 sub-categories. Our criteria offer a foundation for operationalizing and testing processes to certify fairness, both from the perspective of the auditor and the audited organization.

Summary Background Translating findings from animal models to human disease is essential for dissecting disease mechanisms, developing and testing precise therapeutic strategies. The coronavirus disease 2019 (COVID-19) pandemic has highlighted this need, particularly for models showing disease severity-dependent immune responses. Methods Single-cell transcriptomics (scRNAseq) is well poised to reveal similarities and differences between species at the molecular and cellular level with unprecedented resolution. However, computational methods enabling detailed matching are still scarce. Here, we provide a structured scRNAseq-based approach that we applied to scRNAseq from blood leukocytes originating from humans and hamsters affected with moderate or severe COVID-19. Findings Integration of data from patients with COVID-19 with two hamster models that develop moderate (Syrian hamster, Mesocricetus auratus) or severe (Roborovski hamster, Phodopus roborovskii) disease revealed that most cellular states are shared across species. A neural network-based analysis using variational autoencoders quantified the overall transcriptomic similarity across species and severity levels, showing highest similarity between neutrophils of Roborovski hamsters and patients with severe COVID-19, while Syrian hamsters better matched patients with moderate disease, particularly in classical monocytes. We further used transcriptome-wide differential expression analysis to identify which disease stages and cell types display strongest transcriptional changes. Interpretation Consistently, hamsters’ response to COVID-19 was most similar to humans in monocytes and neutrophils. Disease-linked pathways found in all species specifically related to interferon response or inhibition of viral replication. Analysis of candidate genes and signatures supported the results. Our structured neural network-supported workflow could be applied to other diseases, allowing better identification of suitable animal models with similar pathomechanisms across species. Funding This work was supported by German Federal Ministry of Education and Research, (BMBF) grant IDs: 01ZX1304B, 01ZX1604B, 01ZX1906A, 01ZX1906B, 01KI2124, 01IS18026B and German Research Foundation (DFG) grant IDs: 14933180, 431232613.

The Webis MS MARCO Anchor Text 2022 dataset enriches Version 1 and 2 of the document collection of MS MARCO with anchor text extracted from six Common Crawl snapshots. The six Common Crawl snapshots cover the years 2016 to 2021 (between 1.7-3.4 billion documents each). We sampled 1,000 anchor texts for documents with more than 1,000 anchor texts at random and all anchor texts for documents with less than 1,000 anchor texts (this sampling yields that all anchor text is included for 94% of the documents in Version 1 and 97% of documents for Version 2). Overall, the MS MARCO Anchor Text 2022 dataset enriches 1,703,834 documents for Version 1 and 4,821,244 documents for Version 2 with anchor text. Cleaned versions of the MS MARCO Anchor Text 2022 dataset are available in ir_datasets, Zenodo and Hugging Face. The raw dataset with additional information and all metadata for the extracted anchor texts (roughly 100GB) is available on Hugging Face and files.webis.de. The details of the construction of the Webis MS MARCO Anchor Text 2022 dataset are described in the associated paper. If you use this dataset, please cite @InProceedingsfroebe:2022a, address = Berlin Heidelberg New York, author = Maik Fr``obe and Sebastian G''unther and Maximilian Probst and Martin Potthast and Matthias Hagen, booktitle = Advances in Information Retrieval. 44th European Conference on IR Research (ECIR 2022), editor = Matthias Hagen and Suzan Verberne and Craig Macdonald and Christin Seifert and Krisztian Balog and Kjetil Norvrag and Vinay Setty, month = apr, publisher = Springer, series = Lecture Notes in Computer Science, site = Stavanger, Norway, title = The Power of Anchor Text in the Neural Retrieval Era, year = 2022

This paper compares the accuracy of tail risk forecasts with a focus on including realized skewness and kurtosis in "additive" and "multiplicative" models. Utilizing a panel of 960 US stocks, we conduct diagnostic tests, employ scoring functions, and implement rolling window forecasting to evaluate the performance of Value at Risk (VaR) and Expected Shortfall (ES) forecasts. Additionally, we examine the impact of the window length on forecast accuracy. We propose model specifications that incorporate realized skewness and kurtosis for enhanced precision. Our findings provide insights into the importance of considering skewness and kurtosis in tail risk modeling, contributing to the existing literature and offering practical implications for risk practitioners and researchers.

This paper provides a structured overview of the contents of the Special Issue of the Journal of Multivariate Analysis on “Copula modeling from Abe Sklar to the present day,” along with a brief history of the development of the field.

Shorter development cycles, increasing complexity and cost pressure are driving the need for more efficient development processes. Especially in the field of material development, the long and costly experiments are a major bottleneck. To alleviate this, data driven models, supporting the decision making process, have recently gained popularity. However, such models require a structured representation of the development process to allow an efficient training. In this work, a formalism for deriving an efficient representation of material development processs (MDPs) is proposed, shown exemplary on the development of a high modulus steel (HMS). The formalism is based on the combination of graph based process models and the recently proposed concept of ”flowthings”. This allows to efficiently derive a directed acyclic graph (DAG) representation of the MDP with the acquired data. From this, a database for subsequent training of surrogate models is derived, on which several black box models for the MDP are trained. Best-in-class models are chosen based on the root mean squared error (RMSE) on the test set and substantially used for the inverse optimization of the MDP to maximize the specific modulus while meeting additional design constraints. This showcases the potential of the proposed formalism for expediting the MDP by enabling data driven modeling.

In this work, we demonstrate how Machine Learning (ML) techniques can be employed to externalize the knowledge and time intensive process of material parameter identification. This is done on the example of a recent data driven material model for the non-linear shear behavior of glass fiber reinforced polypropylene (GF/PP) (Gerritzen, 2022). A convolutional neural network (CNN) based model architecture is trained to predict material modeling parameters based on the input of stress–strain-curves. The optimal model architecture and training setup are determined by hyperparameter optimization (HPO). Solely virtual data, generated using the target material model, is used throughout the training and HPO. The final CNN is capable of calculating model parameter combinations from experimental stress–strain-curves which lead to excellent agreement between experimental and associated model curve.

Fiber reinforced plastics (FRP) exhibit strongly non-linear deformation behavior. To capture this in simulations, intricate models with a variety of parameters are typically used. The identification of values for such parameters is highly challenging and requires in depth understanding of the model itself. Machine learning (ML) is a promising approach for alleviating this challenge by directly predicting parameters based on experimental results. So far, this works mostly for purely artificial data. In this work, two approaches to generalize to experimental data are investigated: a sequential approach, leveraging understanding of the constitutive model and a direct, purely data driven approach. This is exemplary carried out for a highly non-linear strain rate dependent constitutive model for the shear behavior of FRP.The sequential model is found to work better on both artificial and experimental data. It is capable of extracting well suited parameters from the artificial data under realistic conditions. For the experimental data, the model performance depends on the composition of the experimental curves, varying between excellently suiting and reasonable predictions. Taking the expert knowledge into account for ML-model training led to far better results than the purely data driven approach. Robustifying the model predictions on experimental data promises further improvement.

Interactive visualization empowers users to actively engage with data. This article introduces interactive visualization’s key features and its applications in data analysis, business, science, and journalism. It also highlights challenges, including scalability, complexity, and user engagement, and discusses how two applications address these issues. Despite these challenges, interactive visualization remains a potent tool for data exploration, fostering accessibility and understanding. This introduction sets the stage for exploring its potential in solving real-world problems.

Data-driven design for cities is improving the quality of everyday life of citizens and optimizes the usage of resources. A new aspect is artificial intelligence, which Smart Cities could greatly benefit from. A central problem for urban designers is the unavailability of data to make relevant decisions. Agent-based simulations enable a view of the dynamic properties of the urban system, generating data in its course. However, the simulation must remain sufficiently simple to remain in the realm of computability. The research question of this paper is: How can we make agents behave more realistically to analyze citizens’ mobility behavior? To solve this problem, we first created a simulated virtual environment, where agents can move freely in a small part of a city, the harbor area in Hamburg, Germany. We assumed that happiness is a crucial motivating factor for the movement of citizens. A survey of 130 citizens provided the weights that govern the simulated environment and the happiness score assignation of places. As an AI method, we then used Reinforcement Learning as a general model and Q-learning as an algorithm to generate a baseline. Through randomly traversing the model environment a baseline was created. We are in the process of enhancing Reinforcement Learning with a Deep Q-Network to make the actors learn. Early experiments show a significant improvement over a tabular Q-learning approach. This paper contributes to the literature of urban planning, and data-driven architectural design. The main contribution is replacing the inefficient search for a global maximum of the happiness function, with an efficient local solution global maximum. This has implications for further research in the generation of synthetic data through simulations.

Elementary Cellular Automata (ECA) are a well-studied computational universe that is, despite its simple configurations, capable of impressive computational variety. Harvesting this computation in a useful way has historically shown itself to be difficult, but if combined with reservoir computing (RC), this becomes much more feasible. Furthermore, RC and ECA enable energy-efficient AI, making the combination a promising concept for Edge AI. In this work, we contrast ECA to substrates of Partially-Local CA (PLCA) and Homogeneous Homogeneous Random Boolean Networks (HHRBN). They are, in comparison, the topological heterogeneous counterparts of ECA. This represents a step from ECA towards more biological-plausible substrates. We analyse these substrates by testing on an RC benchmark (5-bit memory), using Temporal Derrida plots to estimate the sensitivity and assess the defect collapse rate. We …

Elementary Cellular Automata (ECA) are well-studied computational universes capable of impressive computational variety, but harnessing their potential has been challenging. When combined with Reservoir Computing (RC), harnessing this computation becomes feasible, and furthermore enables energy efficient AI. This study compares ECA reservoirs to topological heterogeneous and more biological plausible counterparts of Partially-Local CA (PLCA) and Homogeneous Homogeneous Random Boolean Networks (HHRBN). Using the 5-bit memory benchmark, Temporal Derrida plots and collapse rate, finding are that more disordered topology does not equate to more disordered computation and moreover the evidence suggest this heterogeneity shrinks the critical range.

Since paraphrasing is an ill-defined task, the term ``paraphrasing'' covers text transformation tasks with different characteristics. Consequently, existing paraphrasing studies have applied quite different (explicit and implicit) criteria as to when a pair of texts is to be considered a paraphrase, all of which amount to postulating a certain level of semantic or lexical similarity. In this paper, we conduct a literature review and propose a taxonomy to organize the 25~identified paraphrasing (sub-)tasks. Using classifiers trained to identify the tasks that a given paraphrasing instance fits, we find that the distributions of task-specific instances in the known paraphrase corpora vary substantially. This means that the use of these corpora, without the respective paraphrase conditions being clearly defined (which is the normal case), must lead to incomparable and misleading results.

The comparison of multiple (labeled) graphs with unrelated vertex sets is an important task in diverse areas of applications. Conceptually, it is often closely related to multiple sequence alignments since one aims to determine a correspondence, or more precisely, a multipartite matching between the vertex sets. There, the goal is to match vertices that are similar in terms of labels and local neighborhoods. Alignments of sequences and ordered forests, however, have a second aspect that does not seem to be considered for graph comparison, namely the idea that an alignment is a superobject from which the constituent input objects can be recovered faithfully as well-defined projections. Progressive alignment algorithms are based on the idea of computing multiple alignments as a pairwise alignment of the alignments of two disjoint subsets of the input objects. Our formal framework guarantees that alignments have compositional properties that make alignments of alignments well-defined. The various similarity-based graph matching constructions do not share this property and solve substantially different optimization problems. We demonstrate that optimal multiple graph alignments can be approximated well by means of progressive alignment schemes. The solution of the pairwise alignment problem is reduced formally to computing maximal common induced subgraphs. Similar to the ambiguities arising from consecutive indels, pairwise alignments of graph alignments require the consideration of ambiguous edges that may appear between alignment columns with complementary gap patterns. We report a simple reference implementation in Python/NetworkX intended to serve as starting point for further developments. The computational feasibility of our approach is demonstrated on test sets of small graphs that mimimc in particular applications to molecular graphs.

Satellites provide invaluable data regarding climate and weather patterns on a global scale. Satellite observations are crucial in studies aimed at understanding and monitoring the Earth's atmosphere and surface properties. The forward modeling problem computes radiances or reflectances based on surface and atmospheric state knowledge using radiative transfer modeling (RTM). RTTOV, a widely utilized RTM, serves both operational and research applications. This study presents a machine learning (ML) emulator designed to significantly accelerate the acquisition of satellite observations across 36 spectral channels. The ML model predicted 369,303 points across these 36 channels in less than 14 seconds, showcasing its potential over the much more costly RTTOV emulations. We employ supervised learning to compare the performance of three ML models: Random Forest (RF), Neural Networks (NN), and Convolutional Neural Network (CNN). These models were trained on high-resolution large-eddy simulations in a realistic weather prediction setting over Germany and forward-simulated MODIS-based spectral channels. Our approach achieves strong correlations between predicted and reference data across most of the 36 channels in the test dataset. The $R^2$ metric averaged across the 36 channels is 0.91 for the RF model, 0.93 for the NN model, and 0.90 for the CNN model. The normalized RMSE averaged across the 36 channels is 4.07% for the RF model, 3.69% for the NN model, and 4.35% for the CNN model. We discuss the performance of the models in the validation and combined test dataset, as well as in independent channels and timesteps to understand how robust the model is to different data distributions. Overall, the results highlight the potential of machine learning to replace costly RTM models while maintaining high accuracy in emulating radiance and reflectance. Additionally, we generated visible (RGB) satellite images from the predictions to qualitatively inspect the models' performance.

This article describes the design and evaluation of a virtual field trip on the topic of radioactive waste management research for university education. We created an interactive virtual tour through the Mont Terri underground research laboratory by enhancing the virtual experiment information system, designed for domain experts, with background information, illustrations, tasks, tests, and an improved user interface. To put the tour’s content into context, a conventional introductory presentation on the final disposal of radioactive waste was added. A user study with 22 participants proved a good perceived usability of the virtual tour and the virtual field trip’s ability to transfer knowledge. These results suggest a benefit of employing virtual field trips in geoscientific university courses. In addition, it is conceivable to use the virtual field trip as a tool for science communication in the context of participatory processes during nuclear waste disposal site selection processes.

A numerical modelling strategy for the direct pin pressing process of metallic pins into continuous fibre-reinforced thermoplastic organosheets is developed. The joining process is performed above the thermoplasttextquoterights melting temperature, altering the initial material structure of the composite by fibre rearrangement, which in turn influences the load-bearing capacity of the joint. Therefore, the modelling strategy aims at predicting the resultant material structure after pin pressing. The modelling approach considers both the textile architecture and the process parameters (temperature, tool velocity). A sub-meso modelling framework for the fibres based on a multi-filament approach is used. The interaction between fibres and the thermoplastic melt, as well as the matrix flow, is modelled using the Arbitrary Lagrangian Eulerian method. This allows for the prediction of matrix-rich zones and fibre rearrangement around the pin. The promising results show a good agreement of the resultant material structure in terms of compaction and fibre volume content around the pressed pin. Characteristic parameters show an underestimation of the laminate thickness below the pin. Moreover, an evaluation method for evaluating the orientation changes of the virtual multi-filaments is developed and presented to observe and assess fibre rearrangement and fibre volume content in detail during the numerical process simulation. It can be seen that only fibres around the pin are displaced and not in the whole molten area. Furthermore, it can be observed in detail that the initial position of the fibres in relation to the pin determines whether the fibres are displaced in the in-plane or out-of-plane direction.

Voice Interfaces have become ubiquitous as they can make complex technology more usable and accessible. Current voice interfaces, however, often require the user to learn specific speech commands or sentence patterns to use them. This property does not satisfy usability heuristics and causes current language interfaces to underachieve the naturalness of language interaction. To address this issue, we developed a voice interface that is capable of understanding natural everyday language. The overall objective is to build a German language voice interface for socially assistive robots that can work in public spaces. Therefore, we cannot assume the usertextquoterights prior knowledge or experience. Based on recent advances in deep natural language processing, we have built a voice interface that is not restricted to specific speech commands. To test this voice interface, we conducted a study with 47 participants. Results indicate 93% of the given tasks were solved successfully by the target user group without prior training or experience with the voice interface.

In rectal cancer, where part of the patients undergoes chemoradiotherapy, there is a need for improved pretreatment biomarkers applicable to biopsies. Tumor budding (TB) is a biomarker used in colon cancer, and due to its link to epithelial-mesenchymal transition (EMT), is hypothesized to be a potential marker for therapy resistance. Assessment of the utility of tumor buds in rectal biopsies is challenging due to their rarity. As EMT-related processes are also seen in other morphological features beyond tumor buds, we investigated EMT in tumor tissue including morphological features such as tumor cluster size and fibril-like structures. To do so, we leveraged a cohort of colon cancer whole-slide images and another cohort consisting of rectal cancer biopsies, visualized using hyperplex immunofluorescence to identify tumor and EMT-associated proteins. We built a custom image analysis pipeline to detect and segment tumor buds and other morphological features and correlated them with molecular expression intensities. We found strong correlations of EMT up-regulation and morphological transition states, both at the invasive margin and the tumor center. We furthermore observed a link between morpho-molecular transitions and histological growth patterns, which in turn can inform novel biomarkers. Finally, quantification of these morpho-molecular transition states in rectal biopsies showed their impact on survival after neoadjuvant chemoradiotherapy.Competing Interest StatementThere are no conflicts of interest except for the Authors Cansaran Saygili Demir and Joanna Kowal, who are current or former employees of Lunaphore, which is working on commercializing an automated platform to implement seqIF on standard tissue samples.

As runtime reconfiguration is used in an increasing number of hardware architectures, new simulation and modeling tools are needed to support the developer during the design phases. In this article, a language extension for SystemC is presented, together with a design methodology for the description and simulation of dynamically reconfigurable hardware at different levels of abstraction. The library presented offers a high degree of flexibility in the description of reconfiguration features and their management, while allowing runtime reconfiguration simulation, removal, and replacement of custom modules as well as third-party components throughout the architecture development process. In addition, our approach supports the emerging concept of nested reconfiguration and split regions with a minimal simulation overhead of a maximum of three delta cycles for signal and transaction forwarding, and four delta cycles for the reconfiguration process.

This study advances the utility of synthetic study data in hematology, particularly for Acute Myeloid Leukemia (AML), by facilitating its integration into healthcare systems and research platforms through standardization into the Observational Medical Outcomes Partnership (OMOP) and Fast Healthcare Interoperability Resources (FHIR) formats. In our previous work, we addressed the need for high-quality patient data and used CTAB-GAN+ and Normalizing Flow (NFlow) to synthesize data from 1606 patients across four multicenter AML clinical trials. We published the generated synthetic cohorts, that accurately replicate the distributions of key demographic, laboratory, molecular, and cytogenetic variables, alongside patient outcomes, demonstrating high fidelity and usability. The conversion to the OMOP format opens avenues for comparative observational multi-center research by enabling seamless combination with related OMOP datasets, thereby broadening the scope of AML research. Similarly, standardization into FHIR facilitates further developments of applications, e.g. via the SMART-on-FHIR platform, offering realistic test data. This effort aims to foster a more collaborative research environment and facilitate the development of innovative tools and applications in AML care and research.

The chase is a widely implemented approach to reason with tuple-generating dependencies (tgds), used in data exchange, data integration, and ontology-based query answering. However, it is merely a semi-decision procedure, which may fail to terminate. Many decidable conditions have been proposed for tgds to ensure chase termination, typically by forbidding some kind of "cycle'' in the chase process. We propose a new criterion that explicitly allows some such cycles, and yet ensures termination of the standard chase under reasonable conditions. This leads to new decidable fragments of tgds that are not only syntactically more general but also strictly more expressive than the fragments defined by prior acyclicity conditions. Indeed, while known terminating fragments are restricted to PTime data complexity, our conditions yield decidable languages for any k- ExpTime. We further refine our syntactic conditions to obtain fragments of tgds for which an optimised chase procedure decides query entailment in PSpace or k- ExpSpace, respectively.

Machine learning on large-scale genomic or transcriptomic data is important for many novel health applications. For example, precision medicine tailors medical treatments to patients on the basis of individual biomarkers, cellular and molecular states, etc. However, the data required is sensitive, voluminous, heterogeneous, and typically distributed across locations where dedicated machine learning hardware is not available. Due to privacy and regulatory reasons, it is also problematic to aggregate all data at a trusted third party.Federated learning is a promising solution to this dilemma, because it enables decentralized, collaborative machine learning without exchanging raw data. In this paper, we perform comparative experiments with the federated learning frameworks TensorFlow Federated and Flower. Our test case is the training of disease prognosis and cell type classification models. We train the models with distributed transcriptomic data, considering both data heterogeneity and architectural heterogeneity. We measure model quality, robustness against privacy-enhancing noise, computational performance and resource overhead. Each of the federated learning frameworks has different strengths. However, our experiments confirm that both frameworks can readily build models on transcriptomic data, without transferring personal raw data to a third party with abundant computational resources.

Machine Learning is key in modern data analysis. However, privacy concerns related to sharing sensitive data often hinder the full potential of Machine Learning. One of the popular techniques to ensure privacy is the Private Ag-gregation of Teacher Ensembles (PATE) framework. PATE trains an ensemble of teacher models on private data and transfers the knowledge to a student model, with rigorous privacy guarantees derived using Differential Privacy. So far, PATE has been mainly applied to binary and multi-class classifications. However, such problem types often diverge from real-world scenarios. In medicine, patients often have numerous diseases, and in healthcare multiple treatments may be relevant, not just one. In multi-label problems, the objective is to identify not just a single label, but rather a set of relevant labels that characterizes the data point accurately. This paper explores the intricacies of applying PATE to multi-label learning and identifies the challenges involved. Since the original aggregation algorithms can only handle binary and multi-class problems, we generalize them and call the variants GNThreshold, Confident GNThreshold and Inter-active GNThreshold. We provide a formalized description and privacy analysis. We conduct comparative experiments on image and numeric data sets and evaluate model quality and privacy bounds. Finally, we summarize our findings, analyze challenges and discuss possible optimizations for multi-label PATE. Data and Code Availability We experiment with the publicly available NUS-WIDE data set [6] and multiple synthetically generated data sets [26]. Data description and processing details are in section 5. Our code can be found at https//github.com/BonJamon/MultiLabeIPate

Abstract Transpiration (T), the component of evapotranspiration (ET) controlled by the vegetation, dominates terrestrial ET in many ecosystems; however, estimating it accurately, especially at the global scale, remains a considerable challenge. Existing approaches mostly rely on the relationship between T and photosynthesis, but untangling this relationship is difficult and leads to diverging T estimates. Limited in-situ measurements and the inability to directly measure transpiration from space further complicate the reliable assessment of this crucial process in the terrestrial water cycle. Here, we developed a new hybrid Priestley--Taylor (PT) model combined with an Artificial Neural Network (ANN) using globally available remote sensing and reanalysis data of soil moisture, vapor pressure deficit and windspeed. We also take advantage of the newly released global sap flow measurement network SAPFLUXNET. In the proposed approach, we avoid the parameterization of stomatal conductance by training the ANN on the PT-Coefficient $alpha$, obtained by inverting the PT equation. The results showed that our model framework can estimate T in different forest ecosystems based on few predictors. By utilizing forcings from independent datasets, we eliminate the reliance on in-situ measurements for predicting T. Through upscaling actual observations to a larger scale, this model framework helps alleviate the scarcity of T products. Intercomparison of T with ET partitioning methods based on eddy covariance data, shows high performances (KGE of 0.69 in Europe and 0.60 in North America), slightly improving estimates compared to other models. Analysis of contribution of T to ET across 100 FLUXNET sites result in a global mean of 55.2%. We believe that modelling T independent from the carbon cycle can support our understanding of land-atmosphere feedbacks and climate extremes in future research.

Naturally inspired designs of training environments for reinforcement learning (RL) often suffer from highly skewed encounter probabilities, with a small subset of experiences being encountered frequently, while extreme experiences remain rare. Despite recent algorithmic advancements, research has demonstrated that such environments present significant challenges for reinforcement learning algorithms. In this study, we first demonstrate that traditional designs in training environments for RL-based dynamic obstacle avoidance show extremely unbalanced probabilities for obstacle encounters in a way that high-risk scenarios with multiple threatening obstacles are rare. To address this limitation, we propose a traffic-type-independent training environment that allows us to exert control over the difficulty of obstacle encounter experiences. This allows us to customarily shift obstacle encounter probabilities towards high-risk experiences, which are assessed via two metrics: The number of obstacles involved and an existing collision risk metric. Our findings reveal that shifting the training focus towards higher-risk experiences, from which the agent learns, significantly improves the final performance of the agent. To validate the generalizability of our approach, we designed and evaluated two realistic use cases: a mobile robot and a maritime ship facing the threat of approaching obstacles. In both applications, we observed consistent results, underscoring the broad applicability of our proposed approach across various application contexts and independent of the agent’s dynamics. Furthermore, we introduced Gaussian noise to the sensor signals and incorporated different non-linear obstacle behaviors, which resulted in only marginal performance degradation. This demonstrates the robustness of the trained agent in handling environmental uncertainties.

One of the biggest challenges in the development of learning-driven automated driving technologies remains the handling of uncommon, rare events that may have not been encountered in training. Especially when training a model with real driving data, unusual situations, such as emergency brakings, may be underrepresented, resulting in a model that lacks robustness in rare events. This study focuses on car-following based on reinforcement learning and demonstrates that existing approaches, trained with real driving data, fail to handle safety--critical situations. Since collecting data representing all kinds of possible car-following events, including safety--critical situations, is challenging, we propose a training environment that harnesses stochastic processes to generate diverse and challenging scenarios. Our experiments show that training with real data can lead to models that collide in safety--critical situations, whereas the proposed model exhibits excellent performance and remains accident-free, comfortable, and string-stable even in extreme scenarios, such as full-braking by the leading vehicle. Its robustness is demonstrated by simulating car-following scenarios for various reward function parametrizations and a diverse range of artificial and real leader data that were not included in training and were qualitatively different from the learning data. We further show that conventional reward designs can encourage aggressive behavior when approaching other vehicles. Additionally, we compared the proposed model with classical car-following models and found it to achieve equal or superior results.

Dynamic obstacle avoidance (DOA) is a fundamental challenge for any autonomous vehicle, independent of whether it operates in sea, air, or land. This paper proposes a two-step architecture for handling DOA tasks by combining supervised and reinforcement learning (RL). In the first step, we introduce a data-driven approach to estimate the collision risk (CR) of an obstacle using a recurrent neural network, which is trained in a supervised fashion and offers robustness to non-linear obstacle movements. In the second step, we include these CR estimates into the observation space of an RL agent to increase its situational awareness. We illustrate the power of our two-step approach by training different RL agents in a challenging environment that requires to navigate amid multiple obstacles. The non-linear movements of obstacles are exemplarily modeled based on stochastic processes and periodic patterns, although our architecture is suitable for any obstacle dynamics. The experiments reveal that integrating our CR metrics into the observation space doubles the performance in terms of reward, which is equivalent to halving the number of collisions in the considered environment. We also perform a generalization experiment to validate the proposal in an RL environment based on maritime traffic and real-world vessel trajectory data. Furthermore, we show that the architecture's performance improvement is independent of the applied RL algorithm.

We introduce a lower bounding technique for the min max correlation clustering problem and, based on this technique, a combinatorial 4-approximation algorithm for complete graphs. This improves upon the previous best known approximation guarantees of 5, using a linear program formulation (Kalhan et al., 2019), and 40, for a combinatorial algorithm (Davies et al., 2023). We extend this algorithm by a greedy joining heuristic and show empirically that it improves the state of the art in solution quality and runtime on several benchmark datasets.

Solving the Traveling Salesperson Problem (TSP) remains a persistent challenge, despite its fundamental role in numerous generalized applications in modern contexts. Heuristic solvers address the demand for finding high-quality solutions efficiently. Among these solvers, the Lin-Kernighan-Helsgaun (LKH) heuristic stands out, as it complements the performance of genetic algorithms across a diverse range of problem instances. However, frequent timeouts on challenging instances hinder the practical applicability of the solver. Within this work, we investigate a previously overlooked factor contributing to many timeouts: The use of a fixed candidate set based on a tree structure. Our investigations reveal that candidate sets based on Hamiltonian circuits contain more optimal edges. We thus propose to integrate this promising initialization strategy, in the form of POPMUSIC, within an efficient restart version of LKH. As confirmed by our experimental studies, this refined TSP heuristic is much more efficient – causing fewer timeouts and improving the performance (in terms of penalized average runtime) by an order of magnitude – and thereby challenges the state of the art in TSP solving.

AIMS: Artificial intelligence (AI) provides a powerful tool to extract information from digitised histopathology whole slide images. During the last 5 years, academic and commercial actors have developed new technical solutions for a diverse set of tasks, including tissue segmentation, cell detection, mutation prediction, prognostication and prediction of treatment response. In the light of limited overall resources, it is presently unclear for researchers, practitioners and policymakers which of these topics are stable enough for clinical use in the near future and which topics are still experimental, but worth investing time and effort into. METHODS AND RESULTS: To identify potentially promising applications of AI in pathology, we performed an anonymous online survey of 75 computational pathology domain experts from academia and industry. Participants enrolled in 2021 were queried about their subjective opinion on promising and appealing subfields of computational pathology with a focus upon solid tumours. The results of this survey indicate that the prediction of treatment response directly from routine pathology slides is regarded as the most promising future application. This item was ranked highest in the overall analysis and in subgroups by age and professional background. Furthermore, prediction of genetic alterations, gene expression and survival directly from routine pathology images scored consistently high throughout subgroups. CONCLUSIONS: Together, these data demonstrate a possible direction for the development of computational pathology systems in clinical, academic and industrial research in the near future.

INTRODUCTION: Obtaining real-world data from routine clinical care is of growing interest for scientific research and personalized medicine. Despite the abundance of medical data across various facilities - including hospitals, outpatient clinics, and physician practices - the intersectoral exchange of information remains largely hindered due to differences in data structure, content, and adherence to data protection regulations. In response to this challenge, the Medical Informatics Initiative (MII) was launched in Germany, focusing initially on university hospitals to foster the exchange and utilization of real-world data through the development of standardized methods and tools, including the creation of a common core dataset. Our aim, as part of the Medical Informatics Research Hub in Saxony (MiHUBx), is to extend the MII concepts to non-university healthcare providers in a more seamless manner to enable the exchange of real-world data among intersectoral medical sites.METHODS: We investigated what services are needed to facilitate the provision of harmonized real-world data for cross-site research. On this basis, we designed a Service Platform Prototype that hosts services for data harmonization, adhering to the globally recognized Health Level 7 (HL7) Fast Healthcare Interoperability Resources (FHIR) international standard communication format and the Observational Medical Outcomes Partnership (OMOP) common data model (CDM). Leveraging these standards, we implemented additional services facilitating data utilization, exchange and analysis. Throughout the development phase, we collaborated with an interdisciplinary team of experts from the fields of system administration, software engineering and technology acceptance to ensure that the solution is sustainable and reusable in the long term.RESULTS: We have developed the pre-built packages "ResearchData-to-FHIR," "FHIR-to-OMOP," and "Addons," which provide the services for data harmonization and provision of project-related real-world data in both the FHIR MII Core dataset format (CDS) and the OMOP CDM format as well as utilization and a Service Platform Prototype to streamline data management and use.CONCLUSION: Our development shows a possible approach to extend the MII concepts to non-university healthcare providers to enable cross-site research on real-world data. Our Service Platform Prototype can thus pave the way for intersectoral data sharing, federated analysis, and provision of SMART-on-FHIR applications to support clinical decision making.

INTRODUCTION: Seamless interoperability of ophthalmic clinical data is beneficial for improving patient care and advancing research through the integration of data from various sources. Such consolidation increases the amount of data available, leading to more robust statistical analyses, and improving the accuracy and reliability of artificial intelligence models. However, the lack of consistent, harmonized data formats and meanings (syntactic and semantic interoperability) poses a significant challenge in sharing ophthalmic data.METHODS: The Health Level 7 (HL7) Fast Healthcare Interoperability Resources (FHIR), a standard for the exchange of healthcare data, emerges as a promising solution. To facilitate cross-site data exchange in research, the German Medical Informatics Initiative (MII) has developed a core data set (CDS) based on FHIR.RESULTS: This work investigates the suitability of the MII CDS specifications for exchanging ophthalmic clinical data necessary to train and validate a specific machine learning model designed for predicting visual acuity. In interdisciplinary collaborations, we identified and categorized the required ophthalmic clinical data and explored the possibility of its mapping to FHIR using the MII CDS specifications.DISCUSSION: We found that the current FHIR MII CDS specifications do not completely accommodate the ophthalmic clinical data we investigated, indicating that the creation of an extension module is essential.

BACKGROUND: The coexistence of low muscle mass and high fat mass, two interrelated conditions strongly associated with declining health status, has been characterized by only a few protein biomarkers. High-throughput proteomics enable concurrent measurement of numerous proteins, facilitating the discovery of potentially new biomarkers. METHODS: Data derived from the prospective population-based Cooperative Health Research in the Region of Augsburg S4/FF4 cohort study (median follow-up time: 13.5 years) included 1478 participants (756 men and 722 women) aged 55-74 years in the cross-sectional and 608 participants (315 men and 293 women) in the longitudinal analysis. Appendicular skeletal muscle mass (ASMM) and body fat mass index (BFMI) were determined through bioelectrical impedance analysis at baseline and follow-up. At baseline, 233 plasma proteins were measured using proximity extension assay. We implemented boosting with stability selection to enable false positives-controlled variable selection to identify new protein biomarkers of low muscle mass, high fat mass, and their combination. We evaluated prediction models developed based on group least absolute shrinkage and selection operator (lasso) with 100$times$ bootstrapping by cross-validated area under the curve (AUC) to investigate if proteins increase the prediction accuracy on top of classical risk factors. RESULTS: In the cross-sectional analysis, we identified kallikrein-6, C-C motif chemokine 28 (CCL28), and tissue factor pathway inhibitor as previously unknown biomarkers for muscle mass [association with low ASMM: odds ratio (OR) per 1-SD increase in log2 normalized protein expression values (95% confidence interval (CI)): 1.63 (1.37-1.95), 1.31 (1.14-1.51), 1.24 (1.06-1.45), respectively] and serine protease 27 for fat mass [association with high BFMI: OR (95% CI): 0.73 (0.61-0.86)]. CCL28 and metalloproteinase inhibitor 4 (TIMP4) constituted new biomarkers for the combination of low muscle and high fat mass [association with low ASMM combined with high BFMI: OR (95% CI): 1.32 (1.08-1.61), 1.28 (1.03-1.59), respectively]. Including protein biomarkers selected in $geq$90% of group lasso bootstrap iterations on top of classical risk factors improved the performance of models predicting low ASMM, high BFMI, and their combination [delta AUC (95% CI): 0.16 (0.13-0.20), 0.22 (0.18-0.25), 0.12 (0.08-0.17), respectively]. In the longitudinal analysis, N-terminal prohormone brain natriuretic peptide (NT-proBNP) was the only protein selected for loss in ASMM and loss in ASMM combined with gain in BFMI over 14 years [OR (95% CI): 1.40 (1.10-1.77), 1.60 (1.15-2.24), respectively]. CONCLUSIONS: Proteomic profiling revealed CCL28 and TIMP4 as new biomarkers of low muscle mass combined with high fat mass and NT-proBNP as a key biomarker of loss in muscle mass combined with gain in fat mass. Proteomics enable us to accelerate biomarker discoveries in muscle research.

This article discusses the role of the EU-funded project `CONTESSA – Contemporary teaching skills' (https://contessa-project.eu) in modernizing teacher training methodologies in Cambodia and Sri Lanka. Focusing on the pressing need to adapt to 21st-century teaching and learning paradigms, CONTESSA's design reflects an understanding of the technological disparities across these countries. While acknowledging the resource constraints, the program centers its e-learning curriculum on empowering teachers to enhance lesson planning and teaching quality. The program introduces multiple modules, each targeting specific pedagogical skills ranging from digital teaching, learner-centric approaches, embracing classroom diversity, to understanding gender-based nuances in education. Additionally, the article underscores the significance of training teacher trainers to ensure the program's sustainability, detailing a comprehensive Train-the-Trainer initiative that encompasses face-to-face workshops, tutorials, and a user manual. Derived from a four-year-long project, findings indicate the potential of the CONTESSA program in fostering a shift towards widely accessible teacher training, aiming to equip educators with crucial 21st-century teaching skills and promote holistic student learning. The program's broader vision emphasizes a collaborative community of practice, ensuring the professional growth of educators and impactful learning experiences for students.

The integration of Artificial Intelligence (AI) tools in education represents a significant paradigm shift, holding the promise of enhancing learning experiences and outcomes. This study delves into a pivotal inquiry: What motivational effects does using AI in educational settings exert on student learning? This study, involving 64 students, utilizes a mixed-method approach, including online surveys and focus group discussions, to explore the intricate relationship between AI integration and student motivation in education. It employs robust methodological frameworks such as Grounded Theory and is guided by the Situated Learning Approach and constructivist learning theory.Central findings illuminate AI's potential to foster personalized learning pathways, nurturing the development of critical metacognitive skills like critical thinking and problem-solving. Aligned with the Situated Learning paradigm, AI incorporation infuses educational experiences with contextual relevance, fostering genuine and impactful learning interactions. Analysis reveals six distinct motivational categories: Efficiency, Productivity, Autonomy, Competence Acquisition, Competitive Advantage, and Self-regulation. These categories align with the proposed theoretical model's action phases, from situation analysis to reflective consequences. This study reveals new insights through thorough data analysis, enhancing our understanding of education. Students use AI tools to improve information retrieval, productivity, self-directed learning, and content comprehension, gain competitive advantages, and apply knowledge effectively.This study underscores AI's transformative potential and challenges in shaping personalized, learner-centered education globally. It stresses the importance of understanding motivation-related factors for the effective integration of AI in educational contexts, facilitating the creation of efficient, student-centric learning environments.

The workshop titled "The Dual Impact of AI in Academia: From Deskilling Fears to Upskilling Hopes" focuses on examining the complex role of artificial intelligence in education. It addresses concerns about AI potentially reducing the need for traditional skills while also showcasing how it can enhance educational practices and skill sets. Participants will engage in discussions, case studies, and receive expert advice on integrating AI in a way that benefits teaching and learning. The goal is to provide a nuanced understanding of AI's impact and practical approaches for incorporating it into academia.

In higher education's rapidly digitizing landscape, AI tools lead the charge in revolutionizing learning methods. The Erasmus+ project 'Virtual Interface for Smart Interactions Online' (2020-1-AT01-KA226-HE-092653) introduces an AI mentor to address the shifting demands of digital and remote learning, aiming to reconcile emergent student needs with traditional educational theories.Using mixed methods, the inquiry started with a needs analysis involving 89 participants to understand post-pandemic educational challenges. Additionally, focus groups with 46 participants provided diverse insights into the AI mentor's functionality and impact.This analysis revealed that the AI mentor propels student engagement and facilitates a tailored learning journey. Nonetheless, it unveiled the AI mentor's limitations in uniformly catering to the multifaceted needs prevalent within higher education. Furthermore, a critical examination of the AI mentor's alignment with educational theories, particularly constructivism, disclosed notable discrepancies in mimicking the nuanced judgment and interpersonal engagement intrinsic to human mentorship. Predicated on these insights, the study articulates recommendations for the strategic incorporation of AI tools in higher education, underscoring their potential to augment rather than supplant the indispensable human facets of teaching and mentorship. Strategies advocate for an iterative design process informed by continuous feedback and underscore the importance of interdisciplinary collaboration to ensure AI applications' pedagogical and ethical integrity.Comparing the AI mentor to traditional methods sheds light on its distinct advantages and limitations, offering a comprehensive perspective on its educational role. This abstract provides a concise, academically rigorous overview, laying the groundwork for future AI research in education.

This ARQUS online workshop, organised by the University of Graz, will explore how intelligent technologies can enrich teaching practices and support the development of new competencies, while also addressing concerns that these innovations may reduce the relevance of certain academic skills. Participants will engage in discussions, review case studies, and receive practical guidance on how to integrate AI in ways that enhance both teaching and learning. The aim is to shed light on the influence of AI and provide approaches for its application in educational settings. By the end of the workshop, participants will be able to: Develop strategies for incorporating AI into educational practice to enrich the teaching and learning experience and assess their role in supporting skills development.

The ImageCLEF evaluation campaign was integrated with CLEF (Conference and Labs of the Evaluation Forum) for more than 20 years and represents a Multimedia Retrieval challenge aimed at evaluating the technologies for annotation, indexing, and retrieval of multimodal data. Thus, it provides information access to large data collections in usage scenarios and domains such as medicine, argumentation and content recommendation. ImageCLEF 2024 has four main tasks: (i) a Medical task targeting automatic image captioning for radiology images, synthetic medical images created with Generative Adversarial Networks (GANs), Visual Question Answering and medical image generation based on text input, and multimodal dermatology response generation; (ii) a joint ImageCLEF-Touché task Image Retrieval/Generation for Arguments to convey the premise of an argument, (iii) a Recommending task addressing cultural heritage content-recommendation, and (iv) a joint ImageCLEF-ToPicto task aiming to provide a translation in pictograms from natural language. In 2023, participation increased by 67% with respect to 2022 which reveals its impact on the community.

We propose a novel abstraction of the image segmentation task in the form of a combinatorial optimization problem that we call the multi-separator problem. Feasible solutions indicate for every pixel whether it belongs to a segment or a segment separator, and indicate for pairs of pixels whether or not the pixels belong to the same segment. This is in contrast to the closely related lifted multicut problem, where every pixel is associated with a segment and no pixel explicitly represents a separating structure. While the multi-separator problem is np-hard, we identify two special cases for which it can be solved efficiently. Moreover, we define two local search algorithms for the general case and demonstrate their effectiveness in segmenting simulated volume images of foam cells and filaments.

With climate extremes' rising frequency and intensity, robust analytical tools are crucial to predict their impacts on terrestrial ecosystems. Machine learning techniques show promise but require well-structured, high-quality, and curated analysis-ready datasets. Earth observation datasets comprehensively monitor ecosystem dynamics and responses to climatic extremes, yet the data complexity can challenge the effectiveness of machine learning models. Despite recent progress in deep learning to ecosystem monitoring, there is a need for datasets specifically designed to analyse compound heatwave and drought extreme impact. Here, we introduce the DeepExtremeCubes database, tailored to map around these extremes, focusing on persistent natural vegetation. It comprises over 40,000 spatially sampled small data cubes (i.e. minicubes) globally, with a spatial coverage of 2.5 by 2.5 km. Each minicube includes (i) Sentinel-2 L2A images, (ii) ERA5-Land variables and generated extreme event cube covering 2016 to 2022, and (iii) ancillary land cover and topography maps. The paper aims to (1) streamline data accessibility, structuring, pre-processing, and enhance scientific reproducibility, and (2) facilitate biosphere dynamics forecasting in response to compound extremes.

This paper introduces an artificial neural network (ANN) approach to estimate the autoregressive process AR(1) when the autocorrelation parameter is near one. Traditional ordinary least squares (OLS) estimators suffer from biases in small samples, necessitating various correction methods proposed in the literature. The ANN, trained on simulated data, outperforms these methods due to its nonlinear structure. Unlike competitors requiring simulations for bias corrections based on specific sample sizes, the ANN directly incorporates sample size as input, eliminating the need for repeated simulations. Stability tests involve exploring different ANN architectures and activation functions and robustness to varying distributions of the process innovations. Empirical applications on financial and industrial data highlight significant differences among methods, with ANN estimates suggesting lower persistence than other approaches.

Estimating river flood risks under climate change is challenging, largely due to the interacting and combined influences of various flood-generating drivers. However, a more detailed quantitative analysis of such compounding effects and the implications of their interplay remains underexplored on a large scale. Here, we use explainable machine learning to disentangle compounding effects between drivers and quantify their importance for different flood magnitudes across thousands of catchments worldwide. Our findings demonstrate the ubiquity of compounding effects in many floods. Their importance often increases with flood magnitude, but the strength of this increase varies on the basis of catchment conditions. Traditional flood analysis might underestimate extreme flood hazards in catchments where the contribution of compounding effects strongly varies with flood magnitude. Overall, our study highlights the need to carefully incorporate compounding effects in flood risk assessment to improve estimates of extreme floods.

In oncology, Deep Learning has shown great potential to personalise tasks such as tumour type classification, based on per-patient omics data-sets. Being high dimensional, incorporation of such data in one model is a challenge, often leading to one-dimensional studies and, therefore, information loss. Instead, we first propose relying on non-fixed sets of whole genome or whole exome variant-associated sequences, which can be used for supervised learning of oncology-relevant tasks by our Set Transformer based Deep Neural Network, SetQuence. We optimise this architecture to improve its efficiency. This allows for exploration of not just coding but also non-coding variants, from large datasets. Second, we extend the model to incorporate these representations together with multiple other sources of omics data in a flexible way with SetOmic. Evaluation, using these representations, shows improved robustness and reduced information loss compared to previous approaches, while still being computationally tractable. By means of Explainable Artificial Intelligence methods, our models are able to recapitulate the biological contribution of highly attributed features in the tumours studied. This validation opens the door to novel directions in multi-faceted genome and exome wide biomarker discovery and personalised treatment among other presently clinically relevant tasks.

Excessive tree mortality rates prevail in many regions of the world. Understanding tree mortality dynamics remains elusive as this multifaceted phenomenon is influenced by an interplay of abiotic and biotic factors including, but not limited to, global warming, climate extremes, pests, pathogens, and other environmental stressors. Earth observation satellites, coupled with machine learning, present a promising avenue to unravel map standing dead trees and lay the foundation for explaining the underlying dynamics.However, the lack of globally comprehensive, georeferenced training data spanning various biomes and forest types has hindered the development of a unified global product detailing tree mortality patterns. Present ground-based observations, eg, sourced from national inventories, are often sparse, lack standardization, and spatial specificity. Alternatively, aerial imagery captured via drones or airplanes …

textcopyright 2021 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).Idiosyncrasies in human writing styles make it difficult to develop systems for authorship identification that scale well across individuals. In this year's edition of PAN, the authorship identification track focused on open-set authorship verification, so that systems are applied to unknown documents by previously unseen authors in a new domain. As in the previous year, the sizable materials for this campaign were sampled from English-language fanfiction. The calibration materials handed out to the participants were the same as last year, but a new test set was compiled with authors and fandom domains not present in any of the previous datasets. The general setup of the task did not change, i.e., systems still had to estimate the probability of a pair of documents being authored by the same person. We attracted 13 submissions by 10 international teams, which were compared to three complementary baselines, using five diverse evaluation metrics. Post-hoc analyses show that systems benefitted from the abundant calibration materials and were well-equipped to handle the open-set scenario: Both the top-performing approach and the highly competitive cohort of runner-ups presented surprisingly strong verifiers. We conclude that, at least within this specific text variety, (large-scale) open-set authorship verification is not necessarily or inherently more difficult than a closed-set setup, which offers encouraging perspectives for the future of the field.

In today's data-centric world, where data fuels numerous application domains, with machine learning at the forefront, handling the enormous volume of data efficiently in terms of time and energy presents a formidable challenge. Conventional computing systems and accelerators are continually being pushed to their limits to stay competitive. In this context, computing near-memory (CNM) and computing-in-memory (CIM) have emerged as potentially game-changing paradigms. This survey introduces the basics of CNM and CIM architectures, including their underlying technologies and working principles. We focus particularly on CIM and CNM architectures that have either been prototyped or commercialized. While surveying the evolving CIM and CNM landscape in academia and industry, we discuss the potential benefits in terms of performance, energy, and cost, along with the challenges associated with these cutting-edge computing paradigms.

Decision-making and opinion-forming are everyday tasks that involve weighing pro and con arguments. The goal of Touché is to foster the development of support-technologies for decision-making and opinion-forming and to improve our understanding of these processes. This fifth edition of the lab features three shared tasks: (1) Human value detection (ValueEval), where participants detect (implicit) references to human values and their attainment in text; (2) Multilingual Ideology and Power Identification in Parliamentary Debates, where participants identify from a speech the political leaning of the speaker’s party and whether it was governing at the time of the speech (new task); and (3) Image retrieval or generation in order to convey the premise of an argument with visually. In this paper, we briefly describe the planned setup for the fifth lab edition at CLEF 2024 and summarize the results of the 2023 edition.

Climate change elevates the threat of compound heat and drought events, with their ecological and socioeconomic impacts exacerbated by human ecosystem alterations such as eutrophication, salinization, and river engineering. Here, we study how multiple stressors produced an environmental disaster in a large European river, the Oder River, where a toxic bloom of the brackish-water planktonic haptophyte Prymnesium parvum (the ``golden algae'') killed approximately 1000 metric tons of fish and most mussels and snails. We uncovered the complexity of this event using hydroclimatic data, remote sensing, cell counts, hydrochemical and toxin analyses, and genetics. After incubation in impounded upstream channels with drastically elevated concentrations of salts and nutrients, only a critical combination of chronic salt and nutrient pollution, acute high water temperatures, and low river discharge during a heatwave enabled the riverine mass proliferation of B-type P. parvum along a 500 km river section. The dramatic losses of large filter feeders and the spreading of vegetative cells and resting stages make the system more susceptible to new harmful algal blooms. Our findings show that global warming, water use intensification, and chronic ecosystem pollution could increase likelihood and severity of such compound ecoclimatic events, necessitating consideration in future impact models.

Russian Learner Corpus (RLC) is a large collection of learner texts in Russian written by native speakers of over forty languages. Learner errors in part of the corpus are manually corrected and annotated. Diverging from conventional error classifications, which typically focus on isolated lexical and grammatical features, the RLC error classification intends to highlight learners' strategies employed in the process of text production, such as derivational patterns and syntactic relations (including agreement and government). In this paper, we present two open datasets derived from RLC: a manually annotated full-text dataset and a dataset with crowdsourced corrections for individual sentences. In addition, we introduce an automatic error annotation tool that, given an original sentence and its correction, locates and labels errors according to a simplified version of the RLC error-type system. We evaluate the performance of the tool on manually annotated data from RLC.

Most car-following models were originally developed for lane-based traffic. Over the past two decades, efforts have been made to calibrate car-following models for non-lane-based traffic. However, traffic conditions with varying vehicle dimensions, intermittent following, and multiple leaders often occur and make subjective Leader-Follower (LF) pair identification challenging. In this study, we analyze Vehicle Following (VF) behavior in traffic with a lack of lane discipline using high-resolution microscopic trajectory data collected in Chennai, India. The paper's main contributions are threefold. Firstly, three criteria are used to identify LF pairs from the driver's perspective, taking into account the intermittent following, lack of lane discipline due to consideration of lateral separation, and the presence of in-between vehicles. Second, the psycho-physical concept of the regime in the Wiedemann-99 model is leveraged to determine the traffic-dependent "influence zone" for LF identification. Third, a joint and consistent framework is proposed for identifying LF pairs and estimating VF parameters. The proposed methodology outperforms other heuristic-based LF identification methods from the literature in terms of quantitative and qualitative performance measures. The proposed approach can enable robust and more realistic LF identification and VF parameter calibration with practical applications such as LOS analysis, capacity, and travel time estimation.

Electrolysis stands as a pivotal method for environmentally sustainable hydrogen production. However, the formation of gas bubbles during the electrolysis process poses significant challenges by impeding the electrochemical reactions, diminishing cell efficiency, and dramatically increasing energy consumption. Furthermore, the inherent difficulty in detecting these bubbles arises from the non-transparency of the wall of electrolysis cells. Additionally, these gas bubbles induce alterations in the conductivity of the electrolyte, leading to corresponding fluctuations in the magnetic flux density outside of the electrolysis cell, which can be measured by externally placed magnetic sensors. By solving the inverse problem of the Biot–Savart Law, we can estimate the conductivity distribution as well as the void fraction within the cell. In this work, we study different approaches to solve the inverse problem including Invertible Neural Networks (INNs) and Tikhonov regularization. Our experiments demonstrate that INNs are much more robust to solving the inverse problem than Tikhonov regularization when the level of noise in the magnetic flux density measurements is not known or changes over space and time.

Machine Learning (ML) is crucial in many sectors, including computer vision. However, ML models trained on sensitive data face security challenges, as they can be attacked and leak information. Privacy-Preserving Machine Learning (PPML) addresses this by using Differential Privacy (DP) to balance utility and privacy. This study identifies image dataset characteristics that affect the utility and vulnerability of private and non-private Convolutional Neural Network (CNN) models. Through analyzing multiple datasets and privacy budgets, we find that imbalanced datasets increase vulnerability in minority classes, but DP mitigates this issue. Datasets with fewer classes improve both model utility and privacy, while high entropy or low Fisher Discriminant Ratio (FDR) datasets deteriorate the utility-privacy trade-off. These insights offer valuable guidance for practitioners and researchers in estimating and optimizing the utility-privacy trade-off in image datasets, helping to inform data and privacy modifications for better outcomes based on dataset characteristics.

The SciQA benchmark for scientific question answering aims to represent a challenging task for next-generation question-answering systems on which vanilla large language models fail. In this article, we provide an analysis of the performance of language models on this benchmark including prompting and fine-tuning techniques to adapt them to the SciQA task. We show that both fine-tuning and prompting techniques with intelligent few-shot selection allow us to obtain excellent results on the SciQA benchmark. We discuss the valuable lessons and common error categories, and outline their implications on how to optimise large language models for question answering over knowledge graphs.

Most existing computational tools for assumption-based argumentation (ABA) focus on so-called flat frameworks, disregarding the more general case. In this paper, we study an instantiation-based approach for reasoning in possibly non-flat ABA. We make use of a semantics-preserving translation between ABA and bipolar argumentation frameworks (BAFs). By utilizing compilability theory, we establish that the constructed BAFs will in general be of exponential size. In order to keep the number of arguments and computational cost low, we present three ways of identifying redundant arguments. Moreover, we identify fragments of ABA which admit a poly-sized instantiation. We propose two algorithmic approaches for reasoning in possibly non-flat ABA. The first approach utilizes the BAF instantiation while the second works directly without constructing arguments. An empirical evaluation shows that the former outperforms the latter on many instances, reflecting the lower complexity of BAF reasoning. This result is in contrast to flat ABA, where direct approaches dominate instantiation-based approaches.

Clouds are a crucial regulator in the Earth's energy budget through their radiative properties, both at the top of the atmosphere and at the surface; hence, determining key factors like their vertical extent is of essential interest. While the cloud top height is commonly retrieved by satellites, the cloud base height is difficult to estimate from satellite remote sensing data. Here, we present a novel method called ORABase (Ordinal Regression Auto-encoding of cloud Base), leveraging spatially resolved cloud properties from the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument to retrieve the cloud base height over marine areas. A machine learning model is built with two components to facilitate the cloud base height retrieval: the first component is an auto-encoder designed to learn a representation of the data cubes of cloud properties and to reduce their dimensionality. The second component is developed for predicting the cloud base using ground-based ceilometer observations from the lower-dimensional encodings generated by the aforementioned auto-encoder. The method is then evaluated based on a collection of collocated surface ceilometer observations and retrievals from the CALIOP satellite lidar. The statistical model performs similarly on both datasets and performs notably well on the test set of ceilometer cloud bases, where it exhibits accurate predictions, particularly for lower cloud bases, and a narrow distribution of the absolute error, namely 379 and 328 m for the mean absolute error and the standard deviation of the absolute error, respectively. Furthermore, cloud base height predictions are generated for an …

Clouds constitute, through their interactions with incoming solar radiation and outgoing terrestrial radiation, a fundamental element of the Earth’s climate system. Different cloud types show a wide variety in cloud microphysical or optical properties, phase, vertical extent or temperature among others, and thus disparate radiative effects. Both in observational and model datasets, classifying cloud types is also of large importance since different cloud types respond differently to current and future anthropogenic climate change. Cloud types have traditionally been defined using a simplified partition of the space determined by spatially aggregated values e.g. of the cloud top pressure and the cloud optical thickness. In this study, we present a method called CloudViT (Cloud Vision Transformer) building upon spatial extracts of cloud properties from the MODIS instrument to derive cloud types, leveraging spatial features and patterns with a vision transformer model. The classification model is based on global surface observations of cloud types. The method is then evaluated through the distributions of cloud type properties and the corresponding spatial patterns of cloud type occurrences for a global cloud type dataset produced over a year-long period. Subsequently, a first application of the cloud type classification method to climate model data is presented. This application additionally provides insights into how global storm-resolving models are representing clouds as these models are increasingly being used to perform simulations. The global cloud type dataset and the method code constituting CloudViT are available from Zenodo (Lenhardt et al …

Clouds are key regulators of the Earth’s energy budget. Their microphysical and optical properties lead to vastly disparate radiative properties. Retrieving information about clouds is thus crucial to reduce uncerntainties in our estimation of climate change. In this study, we present a common approach to the retrieval of cloud type and cloud base height (CBH), two useful aspects to characterise clouds and their radiative effects.We leverage surface observations of these two cloud characterictics from the network made available by the UK Met Office, linked to satellite retrievals of relevant cloud properties from the MODIS instrument, namely cloud top height, cloud optical thickness and cloud water path. Our approach relies on a convolutional auto-encoder (AE) to project a data cube (dimension of 3 channels, 128 km, 128 km), comprised of the aforementioned cloud properties, to a latent space of lower dimensionality …

While engaging with the unfolding revolution in autonomous driving, a challenge presents itself, how can we effectively raise awareness within society about this transformative trend? While full-scale autonomous driving vehicles often come with a hefty price tag, the emergence of small-scale car platforms offers a compelling alternative. These platforms not only serve as valuable educational tools for the broader public and young generations but also function as robust research platforms, contributing significantly to the ongoing advancements in autonomous driving technology. This survey outlines various small-scale car platforms, categorizing them and detailing the research advancements accomplished through their usage. The conclusion provides proposals for promising future directions in the field.

Autonomous driving presents unique challenges, particularly in transferring agents trained in simulation to real-world environments due to the discrepancies between the two. To address this issue, here we propose a robust Deep Reinforcement Learning (DRL) framework that incorporates platform-dependent perception modules to extract task-relevant information, enabling the training of a lane-following and overtaking agent in simulation. This framework facilitates the efficient transfer of the DRL agent to new simulated environments and the real world with minimal adjustments. We assess the performance of the agent across various driving scenarios in both simulation and the real world, comparing it to human drivers and a proportional-integral-derivative (PID) baseline in simulation. Additionally, we contrast it with other DRL baselines to clarify the rationale behind choosing this framework. Our proposed approach helps bridge the gaps between different platforms and the Simulation to Reality (Sim2Real) gap, allowing the trained agent to perform consistently in both simulation and real-world scenarios, effectively driving the vehicle.

G protein-coupled receptors (GPCRs) orchestrate many physiological functions and are a crucial target in drug discovery. Adhesion GPCRs (aGPCRs), the second largest family within this superfamily, are promising yet underexplored targets for treating various diseases, including obesity, psychiatric disorders, and cancer. However, the receptors' unique and complex structure and miscellaneous interactions complicate comprehensive pharmacological studies. Despite recent progress in determining structures and elucidation of the activation mechanism, the function of many receptors remains to be determined. This review consolidates current knowledge on aGPCR ligands, focusing on small molecule orthosteric ligands and allosteric modulators identified for the ADGRGs subfamily (subfamily VIII), (GPR56/ADGRG1, GPR64/ADGRG2, GPR97/ADGRG3, GPR114/ADGRG5, GPR126/ADGRG6, and GPR128/ADGRG7). We discuss challenges in hit identification, target validation, and drug discovery, highlighting molecular compositions and recent structural breakthroughs. ADGRG ligands can offer new insights into aGPCR modulation and have significant potential for novel therapeutic interventions targeting various diseases.

Urinary tract infection (UTI) is a common disorder. Its diagnosis can be made by microscopic examination of voided urine for markers of infection. This manual technique is technically difficult, time-consuming and prone to inter-observer errors. The application of computer vision to this domain has been slow due to the lack of a clinical image dataset from UTI patients. We present an open dataset containing 300 images and 3,562 manually annotated urinary cells labelled into seven classes of clinically significant cell types. It is an enriched dataset acquired from the unstained and untreated urine of patients with symptomatic UTI using a simple imaging system. We demonstrate that this dataset can be used to train a Patch U-Net, a novel deep learning architecture with a random patch generator to recognise urinary cells. Our hope is, with this dataset, UTI diagnosis will be made possible in nearly all clinical settings by using a simple imaging system which leverages advanced machine learning techniques.

The fusion of traditional chemical descriptors with Graph Neural Networks (GNNs) offers a compelling strategy for enhancing ligand-based virtual screening methodologies. A comprehensive evaluation revealed that the benefits derived from this integrative strategy vary significantly among different GNNs. Specifically, while GCN and SchNet demonstrate pronounced improvements by incorporating descriptors, SphereNet exhibits only marginal enhancement. Intriguingly, despite SphereNettextquoterights modest gain, all three models-GCN, SchNet, and SphereNet-achieve comparable performance levels when leveraging this combination strategy. This observation underscores a pivotal insight: sophisticated GNN architectures may be substituted with simpler counterparts without sacrificing efficacy, provided that they are augmented with descriptors. Furthermore, our analysis reveals a set of expert-crafted descriptorstextquoteright robustness in scaffold-split scenarios, frequently outperforming the combined GNN-descriptor models. Given the critical importance of scaffold splitting in accurately mimicking real-world drug discovery contexts, this finding accentuates an imperative for GNN researchers to innovate models that can adeptly navigate and predict within such frameworks. Our work not only validates the potential of integrating descriptors with GNNs in advancing ligand-based virtual screening but also illuminates pathways for future enhancements in model development and application. Our implementation can be found at https://github.com/meilerlab/gnn-descriptor.Competing Interest StatementThe authors have declared no competing interest.

In this review, we analyze the current state of the art of computational models for in-vehicle User Interface (UI) design. Driver distraction, often caused by drivers performing Non Driving Related Tasks (NDRTs), is a major contributor to vehicle crashes. Accordingly, in-vehicle User Interfaces (UIs) must be evaluated for their distraction potential. Computational models are a promising solution to automate this evaluation, but are not yet widely used, limiting their real-world impact. We systematically review the existing literature on computational models for NDRTs to analyze why current approaches have not yet found their way into practice. We found that while many models are intended for UI evaluation, they focus on small and isolated phenomena that are disconnected from the needs of automotive UI designers. In addition, very few approaches make predictions detailed enough to inform current design processes. Our analysis of the state of the art, the identified research gaps, and the formulated research potentials can guide researchers and practitioners toward computational models that improve the automotive User Interface (UI) design process.

As machine learning-enabled Text-to-Image (TTI) systems are becoming increasingly prevalent and seeing growing adoption as commercial services, characterizing the social biases they exhibit is a necessary first step to lowering their risk of discriminatory outcomes. This evaluation, however, is made more difficult by the synthetic nature of these systems' outputs: common definitions of diversity are grounded in social categories of people living in the world, whereas the artificial depictions of fictive humans created by these systems have no inherent gender or ethnicity. To address this need, we propose a new method for exploring the social biases in TTI systems. Our approach relies on characterizing the variation in generated images triggered by enumerating gender and ethnicity markers in the prompts, and comparing it to the variation engendered by spanning different professions. This allows us to (1) identify specific bias trends, (2) provide targeted scores to directly compare models in terms of diversity and representation, and (3) jointly model interdependent social variables to support a multidimensional analysis. We leverage this method to analyze images generated by 3 popular TTI systems (Dall E 2, Stable Diffusion v 1.4 and 2) and find that while all of their outputs show correlations with US labor demographics, they also consistently under-represent marginalized identities to different extents. We also release the datasets and low-code interactive bias exploration platforms developed for this work, as well as the necessary tools to similarly evaluate additional TTI systems.

Visual analytics (VA) is a paradigm for insight generation by using visual analysis techniques and automated reasoning by transforming data into hypotheses and visualization to extract new insights. The insights are fed back into the data to enhance it until the desired insight is found. Many applications use this principle to provide meaningful mechanisms to assist decision-makers in achieving their goals. This process can be affected by various uncertainties that can interfere with the user decision-making process. Currently, there are no methodical description and handling tool to include uncertainty in VA systematically. We provide a unified workflow to transform the classic VA cycle into an uncertainty-aware visual analytics (UAVA) cycle consisting of five steps. To prove its usability, three real-world applications represent examples of the UAVA cycle implementation and the described workflow.

Chatbots based on large language models are becoming an important tool in modern software development, yet little is known about how programming beginners interact with this new technology to write code and acquire new knowledge. Thus, we are missing key ingredients to develop guidelines on how to adopt chatbots for becoming productive at programming.

Understanding Earth's terrestrial biosphere dynamics is vital for comprehending our planet's environmental health and sustainability. Recently, the frequency and intensity of extreme climate events have risen, significantly impacting the biosphere. Given the advancements of recurrent neural networks in modeling complex, nonlinear dynamics, we explore the capability of recurrent neural network models to model and predict the impacts of extreme events on biosphere dynamics. In this work, we compare four different recurrent network architectures, each with distinct features: 1) Recurrent Neural Networks (RNNs); 2) Long Short-Term Memory-based networks (LSTMs), known for their efficacy in handling long-term dependencies; 3) Gated Recurrent Unit-based networks (GRUs), which offer a simplified yet effective alternative to LSTMs; and 4) Echo State Networks (ESNs), which are distinguished by fixed internal …

The spectral signatures of vegetation are indicative of ecosystem states and health. Spectral indices used to monitor vegetation are characterized by long-term trends, seasonal fluctuations, and responses to weather anomalies. This study investigates the potential of neural networks in learning and predicting vegetation response, including extreme behavior from meteorological data. While machine learning methods, particularly neural networks, have significantly advanced in modeling nonlinear dynamics, it has become standard practice to approach the problem using recurrent architectures capable of capturing nonlinear effects and accommodating both long- and short-term memory. We compare four recurrent-based learning models, which differ in their training and architecture for predicting spectral indices at different forest sites in Europe: (1) recurrent neural networks (RNNs), (2) long short-term memory networks (LSTMs), (3) gated recurrent unit networks (GRUs), and (4) echo state networks (ESNs). While our results show minimal quantitative differences in their performances, ESNs exhibit slightly superior results across various metrics. Overall, we show that recurrent network architectures prove generally suitable for vegetation state prediction yet exhibit limitations under extreme conditions. This study highlights the potential of recurrent network architectures for vegetation state prediction, emphasizing the need for further research to address limitations in modeling extreme conditions within ecosystem dynamics.

Remote sensing is an essential technology in environmental science to study Earth surface processes. In optical remote sensing, spectral indices (SI) are widely used to quantify the properties of specific surface characteristics. SI mathematically combine reflectance values measured at different wavelengths. To gain an overview and access to such indices, comprehensive catalogs have been published and implemented in various programming languages. However, there is no Julia-based tool available for efficiently managing and using these indices. Here we introduce SpectralIndices.jl, a Julia package designed to retrieve and compute SI. Built on the Awesome Spectral Indices (ASI) catalog, our package enables rapid computation of SI using native functions. The multiple dispatch capability of Julia optimizes data handling across various storage types, ensuring quick load times. While primarily based on the ASI collection, SpectralIndices.jl also accommodates custom-made indices, offering users the flexibility to explore and compare alternative indices. The software is open source and available on github.com/awesome-spectral-indices/SpectralIndices.jl

Since students are familiar with machine learning (ML)-based applications in their everyday lives, they already construct mental models of how these systems work. This can result in misconceptions that influence the learning of correct ML concepts. Therefore, this study investigates the misconceptions students hold about the functionality of ML-based applications. To this end, we conducted semi-structured interviews with five students, focusing on their understanding of facial recognition and ChatGPT. The interviews were analyzed using an inductively developed code system and qualitative content analysis. This process identified six key misconceptions held by students: “Programmed Behavior,” “Exactness,” “Data Storage,” “Continuous Learning,” “User-trained Model,” and “Autonomous Data Acquisition”. These misconceptions include the notion that AI learns continuously during application, or that training data is saved and reused later. This paper presents the identified misconceptions and discusses their implication for the design and evaluation of effective learning activities in the context of ML.

The rapid development of vaccines to combat the spread of COVID-19, caused by the SARS-CoV-2 virus, is a great scientific achievement. Before the development of the COVID-19 vaccines, most studies capitalized on the available data that did not include pharmaceutical measures. Such studies focused on the impact of non-pharmaceutical measures such as social distancing, sanitation, use of face masks, and lockdowns to study the spread of COVID-19. In this study, we used the SIDARTHE-V model, an extension of the SIDARTHE model, which includes vaccination rollouts. We studied the impact of vaccination on the severity of the virus, specifically focusing on death rates, in African countries. The SIRDATHE-V model parameters were extracted by simultaneously fitting the COVID-19 cumulative data of deaths, recoveries, active cases, and full vaccinations reported by the governments of Ghana, Kenya, Mozambique, Nigeria, South Africa, Togo, and Zambia. Using South Africa as a case study, our analysis showed that the cumulative death rates declined drastically with the increased extent of vaccination drives. Whilst the infection rates sometimes increased with the arrival of new coronavirus variants, the death rates did not increase as they did before vaccination.

Excited-state molecular dynamics simulations are crucial for understanding processes like photosynthesis, vision, and radiation damage. However, the computational complexity of quantum chemical calculations restricts their scope. Machine learning offers a solution by delivering high-accuracy properties at lower computational costs. We present SpaiNN, an open-source Python software for ML-driven surface hopping nonadiabatic molecular dynamics simulations. SpaiNN combines the invariant and equivariant neural network architectures of SchNetPack with SHARC for surface hopping dynamics. Its modular design allows users to implement and adapt modules easily. We compare rotationally-invariant and equivariant representations in fitting potential energy surfaces of multiple electronic states and properties arising from the interaction of two electronic states. Simulations of the methyleneimmonium cation and various alkenes demonstrate the superior performance of equivariant SpaiNN models, improving accuracy, generalization, and efficiency in both training and inference.

Heterogeneous radiative forcing in mid-latitudes, such as that exerted by aerosols, has been found to affect the Arctic climate, though the mechanisms remain debated. In this study, we leverage Deep Learning (DL) techniques to explore the complex response of the Arctic climate system to local radiative forcing over Europe. We conducted sensitivity experiments using the Max Planck Institute Earth System Model (MPI-ESM1.2) coupled with atmosphere-ocean–land surface components. Utilizing a DL-based clustering method, we classify atmospheric circulation patterns in a lower-dimensional space, focusing on Poleward Moist Static Energy Transport (PMSET) as our primary parameter. We developed a novel method to analyze the circulation patterns' contributions to various climatic parameter anomalies. Our findings indicate that the negative forcing over Europe alters existing circulation patterns and their …

Heterogeneous radiative forcing in mid-latitudes, such as that exerted by aerosols, has been found to affect the Arctic climate, though the mechanisms remain debated. In this study, we leverage deep learning (DL) techniques to explore the complex response of the Arctic climate system to local radiative forcing over Europe. We conducted sensitivity experiments using the Max Planck Institute Earth System Model (MPI-ESM1.2) coupled with atmosphere–ocean–land-surface components. Large-scale circulation patterns can mediate the impact of the forcing on Arctic climate dynamics. We employed a DL-based clustering approach to classify large-scale atmospheric circulation patterns. To enhance the analysis of how these patterns impact the Arctic climate, the poleward moist static energy transport (PMSET) associated with the atmospheric circulation patterns was incorporated as an additional similarity metric in the clustering process. Furthermore, we developed a novel method to analyze the circulation patterns' contributions to various climatic parameter anomalies. Our findings indicate that the negative radiative forcing over Europe alters existing circulation patterns and their occurrence frequency without introducing new ones. Specifically, our analysis revealed that while the regional radiative forcing alters the occurrence frequencies of the circulation patterns, these changes are not the primary drivers of the forcing's impact on the Arctic parameters. Instead, it is the shifts in the mean spatial characteristics of the atmospheric circulation patterns, induced by the forcing, that predominantly determine the effects on the Arctic climate. Our …

Heterogeneous radiative forcing in mid-latitudes, such as those caused by aerosols, has been observed to influence the Arctic climate, although the underlying mechanisms continue to be a subject of scientific discussion. In this research, we employed Deep Learning (DL) methodologies to explore the complex response of the Arctic climate system to localized radiative forcing over Europe. We performed sensitivity experiments using the Max Planck Institute Earth System Model (MPI-ESM1. 2). By applying a DL-driven clustering approach, we classified atmospheric circulation patterns within a reduced-dimensional framework, with a particular focus on Poleward Moist Static Energy Transport (PMSET) as our primary parameter of interest. Additionally, we developed a new methodology to assess the contributions of these circulation patterns to anomalies in various climatic parameters.Our results demonstrate that …

Each year the International Semantic Web Conference accepts a set of Semantic Web Challenges to establish competitions that will advance the state of the art solutions in any given problem domain. The SeMantic AnsweR Type prediction task (SMART) was part of ISWC 2020 challenges. Question type and answer type prediction can play a key role in knowledge base question answering systems providing insights that are helpful to generate correct queries or rank the answer candidates. More concretely, given a question in natural language, the task of SMART challenge is, to predict the answer type using a target ontology (e.g., DBpedia or Wikidata).

Recent advancements in Earth system science have been marked by the exponential increase in the availability of diverse, multivariate datasets characterised by moderate to high spatio-temporal resolutions. Earth System Data Cubes (ESDCs) have emerged as one suitable solution for transforming this flood of data into a simple yet robust data structure. ESDCs achieve this by organising data into an analysis-ready format aligned with a spatio-temporal grid, facilitating user-friendly analysis and diminishing the need for extensive technical data processing knowledge. Despite these significant benefits, the completion of the entire ESDC life cycle remains a challenging task. Obstacles are not only of a technical nature but also relate to domain-specific problems in Earth system research. There exist barriers to realising the full potential of data collections in light of novel cloud-based technologies, particularly in curating data tailored for specific application domains. These include transforming data to conform to a spatio-temporal grid with minimum distortions and managing complexities such as spatio-temporal autocorrelation issues. Addressing these challenges is pivotal for the effective application of Artificial Intelligence (AI) approaches. Furthermore, adhering to open science principles for data dissemination, reproducibility, visualisation, and reuse is crucial for fostering sustainable research. Overcoming these challenges offers a substantial opportunity to advance data-driven Earth system research, unlocking the full potential of an integrated, multidimensional view of Earth system processes. This is particularly true when such research is coupled …

The Sentinel-2 (S2) mission from the European Space Agency's Copernicus program provides essential data for Earth surface analysis. Its Level-2A products deliver high-to-medium resolution (10-60 m) surface reflectance (SR) data through the MultiSpectral Instrument (MSI). To enhance the accuracy and comparability of SR data, adjustments simulating a nadir viewing perspective are essential. These corrections address the anisotropic nature of SR and the variability in sun and observation angles, ensuring consistent image comparisons over time and under different conditions. The -factor method, a simple yet effective algorithm, adjusts observed S2 SR by using the MODIS BRDF model to achieve Nadir BRDF Adjusted Reflectance (NBAR). Despite the straightforward application of the -factor to individual images, a cohesive Python framework for its application across multiple S2 images and Earth System Data Cubes (ESDCs) from cloud-stored data has been lacking. Here we introduce sen2nbar, a Python package crafted to convert S2 SR data to NBAR, supporting both individual images and ESDCs derived from cloud-stored data. This package simplifies the conversion of S2 SR data to NBAR via a single function, organized into modules for efficient process management. By facilitating NBAR conversion for both SAFE files and ESDCs from SpatioTemporal Asset Catalogs (STAC), sen2nbar is developed as a flexible tool that can handle diverse data format requirements. We anticipate that sen2nbar will considerably contribute to the standardization and harmonization of S2 data, offering a robust solution for a diverse range of users across …

Accurate quantification of Gross Primary Production (GPP) is crucial for understanding terrestrial carbon dynamics. It represents the largest atmosphere-to-land CO 2 flux, especially significant for forests. Eddy Covariance (EC) measurements are widely used for ecosystem-scale GPP quantification but are globally sparse. In areas lacking local EC measurements, remote sensing (RS) data are typically utilised to estimate GPP after statistically relating them to in-situ data. Deep learning offers novel perspectives, and the potential of recurrent neural network architectures for estimating daily GPP remains underexplored. This study presents a comparative analysis of three architectures: Recurrent Neural Networks (RNNs), Gated Recurrent Units (GRUs), and Long-Short Term Memory (LSTMs). Our findings reveal comparable performance across all models for full-year and growing season predictions. Notably, LSTMs …

Quantifying Gross Primary Production (GPP) is fundamental for understanding terrestrial carbon dynamics, particularly in forests. The overarching question we address here is whether integrating remote sensing (RS) with deep learning (DL) methodologies can enhance the estimation of daily forest GPP on a European scale.The Eddy Covariance (EC) method, although widely used to infer ecosystem-scale estimates of GPP from in situ CO2 exchange measurements, suffers from limited global coverage. When EC data are not available, RS data are often employed to estimate GPP by establishing statistical relationships with in situ observations. Recently, Machine Learning (ML) strategies, particularly involving RS and meteorological inputs, have been used for estimating GPP continuously in space and time.

Excess tree mortality in the wake of climate extremes has been observed globally. However, we still lack precise data on mortality at global scale to understand respective drivers and spatiotemporal dynamics. The Sentinel-2 satellite fleet, equipped with the MultiSpectral Instrument (MSI), covers the entire earth on average every five days at spatial resolutions ranging from 10 m to 60 m. Mapping tree mortality from Sentinel-2 globally in diverse ecosystems requires equally diverse reference data. Using globally distributed high-resolution aerial orthoimagery reference data and artificial intelligence methods, we can translate spectral signatures of remote sensing into deadwood. Specifically, in this study we show how to predict the share of standing deadwood for a 10 m pixel in a specific year. The method takes into account temporal patterns, spatial context, as well as all Sentinel-2 spectral bands. This will enable us …

Excessive tree mortality is a global concern and remains poorly understood as it is a complex phenomenon. We lack global and temporally continuous coverage on tree mortality data. Ground-based observations on tree mortality, e.g., derived from national inventories, are very sparse, not standardized and not spatially explicit. Earth observation data, combined with supervised machine learning, offer a promising approach to map tree mortality over time. However, global-scale machine learning requires broad training data covering a wide range of environmental settings and forest types. Drones provide a cost-effective source of training data by capturing high-resolution orthophotos of tree mortality events at sub-centimeter resolution. Here, we introduce deadtrees.earth, an open-access platform hosting more than a thousand centimeter-resolution orthophotos, covering already more than 300,000 ha, of which more than 58,000 ha are fully annotated. This community-sourced and rigorously curated dataset shall serve as a foundation for a global initiative to gather comprehensive reference data. In concert with Earth observation data and machine learning it will serve to uncover tree mortality patterns from local to global scales. This will provide the foundation to attribute tree mortality patterns to environmental changes or project tree mortality dynamics to the future. Thus, the open and interactive nature of deadtrees.earth together with the collective effort of the community is meant to continuously increase our capacity to uncover and understand tree mortality patterns.

Volume electron microscopy is the method of choice for the in situ interrogation of cellular ultrastructure at the nanometer scale, and with the increase in large raw image datasets generated, improving computational strategies for image segmentation and spatial analysis is necessary. Here we describe a practical and annotation-efficient pipeline for organelle-specific segmentation, spatial analysis and visualization of large volume electron microscopy datasets using freely available, user-friendly software tools that can be run on a single standard workstation. The procedures are aimed at researchers in the life sciences with modest computational expertise, who use volume electron microscopy and need to generate three-dimensional (3D) segmentation labels for different types of cell organelles while minimizing manual annotation efforts, to analyze the spatial interactions between organelle instances and to visualize the 3D segmentation results. We provide detailed guidelines for choosing well-suited segmentation tools for specific cell organelles, and to bridge compatibility issues between freely available open-source tools, we distribute the critical steps as easily installable Album solutions for deep learning segmentation, spatial analysis and 3D rendering. Our detailed description can serve as a reference for similar projects requiring particular strategies for single- or multiple-organelle analysis, which can be achieved with computational resources commonly available to single-user setups.

Modelling the extremal dependence of bivariate variables is important in a wide variety of practical applications, including environmental planning, catastrophe modelling and hydrology. The majority of these approaches are based on the framework of bivariate regular variation, and a wide range of literature is available for estimating the dependence structure in this setting. However, such procedures are only applicable to variables exhibiting asymptotic dependence, even though asymptotic independence is often observed in practice. In this paper, we consider the so-called textquoteleftangular dependence functiontextquoteright; this quantity summarises the extremal dependence structure for asymptotically independent variables. Until recently, only pointwise estimators of the angular dependence function have been available. We introduce a range of global estimators and compare them to another recently introduced technique for global estimation through a systematic simulation study, and a case study on river flow data from the north of England, UK.

The modelling of multivariate extreme events is important in a wide variety of applications, including flood risk analysis, metocean engineering and financial modelling. A wide variety of statistical techniques have been proposed in the literature; however, many such methods are limited in the forms of dependence they can capture, or make strong parametric assumptions about data structures. In this article, we introduce a novel inference framework for bivariate extremes based on a semi-parametric angular-radial model. This model overcomes the limitations of many existing approaches and provides a unified paradigm for assessing joint tail behaviour. Alongside inferential tools, we also introduce techniques for assessing uncertainty and goodness of fit. Our proposed technique is tested on simulated data sets alongside observed metocean time seriestextquoteright, with results indicating generally good performance.

Buildings remain pivotal in global energy consumption, necessitating a focused approach toward enhancing their energy efficiency to alleviate environmental impacts. Precise energy prediction stands as a linchpin in optimizing efficiency, offering indispensable foresight into future energy demands critical for sustainable environments. However, accurately forecasting energy consumption for individual households and commercial buildings presents multifaceted challenges due to their diverse consumption patterns. Leveraging the emerging landscape of the Internet of Things (IoT) in smart homes, coupled with AI-driven energy solutions, presents promising avenues for overcoming these challenges. This study introduces a pioneering approach that harnesses a hybrid deep learning model for energy consumption prediction, strategically amalgamating convolutional neural networks’ features with long short-term memory (LSTM) units. The model harnesses the granularity of IoT-enabled smart meter data, enabling precise energy consumption forecasts in both residential and commercial spaces. In a comparative analysis against established deep learning models, the proposed hybrid model consistently demonstrates superior performance, notably exceling in accurately predicting weekly average energy usage. The study’s innovation lies in its novel model architecture, showcasing an unprecedented capability to forecast energy consumption patterns. This capability holds significant promise in guiding tailored energy management strategies, thereby fostering optimized energy consumption practices in buildings. The demonstrated superiority of the hybrid model underscores its potential to serve as a cornerstone in driving sustainable energy utilization, offering invaluable guidance for a more energy-efficient future.

We develop a fixed-parameter tractable (FPT) algorithm for skeptical preferred reasoning in assumption-based argumentation (ABA). To this end we make use of so-called backdoors, i.e. sets of assumptions that need to be evaluated s.t. the remaining ABA framework (ABAF) belongs to a computational beneficial sub-class. In order to identify such target classes, we employ a suitable notion of a dependency graph of an ABAF. We show that these graphs can be constructed in polynomial time and that one can efficiently check sufficient properties ensuring that reasoning in the underlying ABAF is tractable. After establishing the theoretical foundations, we test our implementation against the ASPforABA solver which convincingly won the ABA track of the ICCMA'23 competition. As it turns out, our algorithm outperforms ASPforABA on instances with small backdoor sizes.

Despite recent progress, large language models (LLMs) still face the challenge of appropriately reacting to the intricacies of social and cultural conventions. This paper presents Mango, a methodology for distilling high-accuracy, high-recall assertions of cultural knowledge. We judiciously and iteratively prompt LLMs for this purpose from two entry points, concepts and cultures. Outputs are consolidated via clustering and generative summarization. Running the Mango method with GPT-3.5 as underlying LLM yields 167K high-accuracy assertions for 30K concepts and 11K cultures, surpassing prior resources by a large margin in quality and size. In an extrinsic evaluation for intercultural dialogues, we explore augmenting dialogue systems with cultural knowledge assertions. Notably, despite LLMs inherently possessing cultural knowledge, we find that adding knowledge from Mango improves the overall quality, specificity, and cultural sensitivity of dialogue responses, as judged by human annotators. Data and code are available for download.

High-resolution simulations, such as the ICOsahedral Non-hydrostatic Large-Eddy Model (ICON-LEM), provide valuable insights into the complex interactions among aerosols, clouds, and precipitation, which are the major contributors to climate change uncertainty. However, due to their exorbitant computational costs, they can only be employed for a limited period and geographical area. To address this, we propose a more cost-effective method powered by an emerging machine learning approach to better understand the intricate dynamics of the climate system. Our approach involves active learning techniques by leveraging high-resolution climate simulation as an oracle that is queried based on an abundant amount of unlabeled data drawn from satellite observations. In particular, we aim to predict autoconversion rates, a crucial step in precipitation formation, while significantly reducing the need for a large number of labeled instances. In this study, we present novel methods: custom fusion query strategies for labeling instances–weight fusion (WiFi) and merge fusion (MeFi)–along with active feature selection based on SHAP. These methods are designed to tackle real-world challenges–in this case, climate change, with a specific focus on the prediction of autoconversion rates–due to their simplicity and practicality in application.

Precipitation is one of the most relevant weather and climate processes. Its formation rate is sensitive to perturbations such as by the interactions between aerosols, clouds, and precipitation. These interactions constitute one of the biggest uncertainties in determining the radiative forcing of climate change. High-resolution simulations such as the ICOsahedral non-hydrostatic large-eddy model (ICON-LEM) offer valuable insights into these interactions. However, due to exceptionally high computation costs, it can only be employed for a limited period and area. We address this challenge by developing new models powered by emerging machine learning approaches capable of forecasting autoconversion rates—the rate at which small droplets collide and coalesce becoming larger droplets—from satellite observations providing long-term global spatial coverage for more than two decades. In particular, our approach …

Weather can have a significant impact on the power grid. Heat and cold waves lead to increased energy use as customers cool or heat their space, while simultaneously hampering energy production as the environment deviates from ideal operating conditions. Extreme heat has previously melted power cables, while extreme cold can cause vital parts of the energy infrastructure to freeze. Utilities have reserves to compensate for the additional energy use, but in extreme cases which fall outside the forecast energy demand, the impact on the power grid can be severe. In this paper, we present an interactive tool to explore the relationship between weather and power outages. We demonstrate its use with the example of the impact of Winter Storm Uri on Texas in February 2021.

Flood event set generation, as employed in catastrophe risk models, relies on gauge information that is not available in data-scarce regions. To overcome this limitation, we develop a stochastic fluvial and pluvial flood model of Southeast Asia, using freely and globally available discharge data from the global hydrological model GloFAS and rainfall from the ERA5 reanalysis. We use a conditional multivariate statistical model to produce a synthetic catalog of 10,000 years of flood events. We calculate the flood population exposure associated with each flood event using freely available population data from WorldPop and generate exposure probability exceedance curves. We validate the population exposure curves against observed flood disaster data from EM-DAT, showing that our methodology provides exposure estimates that are in line with historical observations. We find that there is a 1% probability that more than 30 million people will be exposed to flooding in a given year according to our event set. This number is roughly half the population living in the 100-year return period flood zone of Fathom's hazard maps, suggesting most studies based on static flood maps overestimate exposure. This analysis provides significant progress over previous non-stochastic studies which are only able to compute total or average exposure within a given floodplain area and demonstrates that a reanalysis-based stochastic flood model can be designed to generate reliable estimates of population exposure probability exceedance. This study is a step toward a fully global catastrophe model for floods capable of providing exposure and loss estimates worldwide.

We provide a mathematically proven parallelization scheme for particle methods on distributed-memory computer systems. Particle methods are a versatile and widely used class of algorithms for computer simulations and numerical predictions in various applications, ranging from continuum fluid dynamics to discrete granular flows and molecular dynamics simulations. Particle methods naturally lend themselves to implementation on parallel computing systems. So far, however, a mathematical proof of correctness and equivalence to sequential implementations was only available for shared-memory parallelism. Here, we leverage a formal definition of the algorithmic class of particle methods to provide a proven parallelization scheme for distributed-memory computers. We prove that thus parallelized particle methods on distributed-memory computers are formally equivalent to their sequential counterpart for a well-defined class of particle methods, and we provide analytical expressions for the speed-up and scalability bounds of this class of algorithms in function of their parameters. The parallelization scheme analyzed here is the basis of many real-world software designs for parallel particle methods. The present analysis is, therefore, of direct relevance to existing and new parallel implementations of particle methods and places them on solid theoretical grounds, rationalizing best practices and providing useful scalability and speedup bounds for benchmarking.

The forecast accuracy of deep-learning-based weather prediction models is improving rapidly, leading many to speak of a "second revolution in weather forecasting". With numerous methods being developed, and limited physical guarantees offered by deep-learning models, there is a critical need for comprehensive evaluation of these emerging techniques. While this need has been partly fulfilled by benchmark datasets, they provide little information on rare and impactful extreme events, or on compound impact metrics, for which model accuracy might degrade due to misrepresented dependencies between variables. To address these issues, we compare deep-learning weather prediction models (GraphCast, PanguWeather, FourCastNet) and ECMWF's high-resolution forecast (HRES) system in three case studies: the 2021 Pacific Northwest heatwave, the 2023 South Asian humid heatwave, and the North American winter storm in 2021. We find evidence that machine learning (ML) weather prediction models can locally achieve similar accuracy to HRES on record-shattering events such as the 2021 Pacific Northwest heatwave and even forecast the compound 2021 North American winter storm substantially better. However, extrapolating to extreme conditions may impact machine learning models more severely than HRES, as evidenced by the comparable or superior spatially- and temporally-aggregated forecast accuracy of HRES for the two heatwaves studied. The ML forecasts also lack variables required to assess the health risks of events such as the 2023 South Asian humid heatwave. Generally, case-study-driven, impact-centric …

This paper develops a Deep Reinforcement Learning (DRL)-based approach for the navigation and control of autonomous surface vessels (ASVs) on inland waterways, where spatial constraints and environmental challenges such as high flow velocities and shallow banks require precise maneuvering. By implementing a state-of-the-art bootstrapped deep Q-learning (DQN) algorithm alongside a novel, flexible training environment generator, we developed a robust and accurate rudder control system capable of adapting to the dynamic conditions of inland waterways. The effectiveness of our approach is validated through comparisons with a vessel-specific Proportional-Integral-Derivative (PID) and standard DQN controller using real-world data from the lower and middle Rhine. It was found that the DRL algorithm demonstrates superior adaptability and generalizability across previously unseen scenarios and achieves high navigational accuracy. Our findings highlight the limitations of traditional control methods like PID in complex river environments, as well as the importance of training in diverse and realistic environments.

Ship trajectories from Automatic Identification System (AIS) messages are important in maritime safety, domain awareness, and algorithmic testing. Although the specifications for transmitting and receiving AIS messages are fixed, it is well known that technical inaccuracies and lacking seafarer compliance lead to severe data quality impairment. This paper proposes an adaptable, data-driven, maneuverability-dependent, α-quantile-based framework for decoding, constructing, splitting, and assessing trajectories from raw AIS records to improve transparency in AIS data mining. Results indicate the proposed filtering algorithm robustly extracts clean, long, and uninterrupted trajectories for further processing. An open-source Python implementation of the framework is provided.

In biological and medical research, scientists now routinely acquire microscopy images of hundreds of morphologically heterogeneous organoids and are then faced with the task of finding patterns in the image collection, i.e., subsets of organoids that appear similar and potentially represent the same morphological class. We adopt models and algorithms for correlating organoid images, i.e., for quantifying the similarity in appearance and geometry of the organoids they depict, and for clustering organoid images by consolidating conflicting correlations. For correlating organoid images, we adopt and compare two alternatives, a partial quadratic assignment problem and a twin network. For clustering organoid images, we employ the correlation clustering problem. Empirically, we learn the parameters of these models, infer a clustering of organoid images, and quantify the accuracy of the inferred clusters, with respect to a training set and a test set we contribute of state-of-the-art light microscopy images of organoids clustered manually by biologists.

Terrestrial surface processes exhibit distinctive spectral signatures captured by optical satellites. Despite the development of over two hundred spectral indices (SIs), current studies often narrow their focus to individual SIs, overlooking the broader context of land surface processes. This project seeks to understand the holistic features of Sentinel-2 based SIs and their relationships with human impact and overall land surface dynamics. To address this, we propose an AI-driven approach that synthesises SIs derived from Sentinel data through dimension reduction, yielding interpretable latent variables describing the system comprehensively. Our goals are to (i) reduce the number of SIs and (ii) compute a few latent variables representing spatio-temporal dynamics, which culminate in a Feature Data Cube. This fully descriptive cube reduces computational costs, facilitating diverse applications. We plan to demonstrate …

The 2015 Paris agreement was established to limit Greenhouse gas (GHG) global warming below 1.5textdegreeC above preindustrial era values. Knowledge of climate sensitivity to GHG levels is central for formulating effective climate policies, yet its exact value is shroud in uncertainty. Climate sensitivity is quantitatively expressed in terms of Equilibrium Climate Sensitivity (ECS) and Transient Climate Response (TCR), estimating global temperature responses after an abrupt or transient doubling of CO2. Here, we represent the complex and highly-dimensional behavior of modelled climate via low-dimensional emergent networks to evaluate Climate Sensitivity (netCS), by first reconstructing meaningful components describing regional subprocesses, and secondly inferring the causal links between these to construct causal networks. We apply this methodology to Sea Surface Temperature (SST) simulations and investigate two different metrics in order to derive weighted estimates that yield likely ranges of ECS (2.35--4.81textdegreeC) and TCR (1.53-2.60textdegreeC). These ranges are narrower than the unconstrained distributions and consistent with the ranges of the IPCC AR6 estimates. More importantly, netCS demonstrates that SST patterns (at ``fast'' timescales) are linked to climate sensitivity; SST patterns over the historical period exclude median sensitivity but not low-sensitivity (ECSthinspace

The management and digitalization of experimental-numerical validation and verification processes within product development (PD) for innovative, lightweight, high-performance composite materials is a major challenge due to increasing process chain complexity, highly heterogeneous data and the need to consider product-related data and processes over their entire life cycle. Methods for modeling the experimental and numerical validation process using Product Lifecycle Management (PLM) concepts to manage and integrate data and Model-Based Systems Engineering (MBSE) approaches to track data links and process steps are elaborated. A process roadmap is created including key processes, subprocesses and metadata for procedures in PD which are subsequently illustrated by an event-driven process chain showing linked processes, inputs/outputs, software needs and iteration points. This is used as a base to model key processes and their properties by creating customizable, connectable blocks in MBSE, where digital data is accessible through an interface linked by respective metadata in a PLMtool. Excerpts of linked processes and respective metadata are exemplarily shown for a hybrid composite structure to illustrate the potential of the elaborated approaches for improved PD efficiency through data reuse, workflow execution, consistent data management and traceability.

In this paper, we address the problem of computing all locally optimal solutions of a given multi-objective problem whose images are sufficiently close to the Pareto front. Such -locally optimal solutions are particularly interesting in the context of multi-objective multimodal optimization (MMO). To accomplish this task, we first define a new set of interest, LQ, that is strongly related to the recently proposed set of -acceptable solutions. Next, we propose a new unbounded archiver, ArchiveUpdateLQ, aiming to capture LQ,in the limit. This archiver can in principle be used in combination with any multi-objective evolutionary algorithm (MOEA). Further, we equip numerous MOEAs with ArchiveUpdateLQ, investigate their performances across several benchmark functions, and compare the enhanced MOEAs with their archive-free counterparts. For our experiments, we utilize the well-established metrics HV, IGDX, and p. Additionally, we propose and use a new performance indicator, IEDR, which results in comparable performances but which is applicable to problems defined in higher dimensions (in particular in decision variable space).

To increase trust in systems, engineers strive to create explanations that are as accurate as possible. However, if the system’s accuracy is compromised, providing explanations for its incorrect behavior may inadvertently lead to misleading explanations. This concern is particularly pertinent when the correctness of the system is difficult for users to judge. In an online survey experiment with 162 participants, we analyze the impact of misleading explanations on users’ perceived and demonstrated trust in a system that performs a hardly assessable task in an unreliable manner. Participants who used a system that provided potentially misleading explanations rated their trust significantly higher than participants who saw the system’s prediction alone. They also aligned their initial prediction with the system’s prediction significantly more often. Our findings underscore the importance of exercising caution when generating explanations, especially in tasks that are inherently difficult to evaluate. The paper and supplementary materials are available at https://doi.org/10.17605/osf.io/azu72

The use of artificial intelligence (AI) in healthcare is transforming a number of medical fields, including nephrology. The integration of various AI techniques in nephrology facilitates the prediction of the early detection, diagnosis, prognosis, and treatment of kidney disease. Nevertheless, recent reports have demonstrated that the majority of published clinical AI studies lack uniform AI reporting standards, which poses significant challenges in interpreting, replicating, and translating the studies into routine clinical use. In response to these issues, worldwide initiatives have created guidelines for publishing AI-related studies that outline the minimal necessary information that researchers should include. By following standardized reporting frameworks, researchers and clinicians can ensure the reproducibility, reliability, and ethical use of AI models. This will ultimately lead to improved research outcomes, enhanced clinical decision-making, and better patient management. This review article highlights the importance of adhering to AI reporting guidelines in medical research, with a focus on nephrology and urology, and clinical practice for advancing the field and optimizing patient care.

Deep-learning models that learn a sense of language on DNA have achieved a high level of performance on genome biological tasks. Genome sequences follow rules similar to natural language but are distinct in the absence of a concept of words. We established byte-pair encoding on the human genome and trained a foundation language model called GROVER (Genome Rules Obtained Via Extracted Representations) with the vocabulary selected via a custom task, next-k-mer prediction. The defined dictionary of tokens in the human genome carries best the information content for GROVER. Analysing learned representations, we observed that trained token embeddings primarily encode information related to frequency, sequence content and length. Some tokens are primarily localized in repeats, whereas the majority widely distribute over the genome. GROVER also learns context and lexical ambiguity. Average trained embeddings of genomic regions relate to functional genomics annotation and thus indicate learning of these structures purely from the contextual relationships of tokens. This highlights the extent of information content encoded by the sequence that can be grasped by GROVER. On fine-tuning tasks addressing genome biology with questions of genome element identification and protein--DNA binding, GROVER exceeds other models' performance. GROVER learns sequence context, a sense for structure and language rules. Extracting this knowledge can be used to compose a grammar book for the code of life.

Transformer-based large language models (LLMs) are very suited for biological sequence data, because of analogies to natural language. Complex relationships can be learned, because a concept of ``words'' can be generated through tokenization. Training the models with masked token prediction, they learn both token sequence identity and larger sequence context. We developed methodology to interrogate model learning, which is both relevant for the interpretability of the model and to evaluate its potential for specific tasks. We used DNABERT, a DNA language model trained on the human genome with overlapping k-mers as tokens. To gain insight into the modeltextasciiacutexs learning, we interrogated how the model performs predictions, extracted token embeddings, and defined a fine-tuning benchmarking task to predict the next tokens of different sizes without overlaps. This task evaluates foundation models without interrogating specific genome biology, it does not depend on tokenization strategies, vocabulary size, the dictionary, or the number of training parameters. Lastly, there is no leakage of information from token identity into the prediction task, which makes it particularly useful to evaluate the learning of sequence context. We discovered that the model with overlapping k-mers struggles to learn larger sequence context. Instead, the learned embeddings largely represent token sequence. Still, good performance is achieved for genome-biology-inspired fine-tuning tasks. Models with overlapping tokens may be used for tasks where a larger sequence context is of less relevance, but the token sequence directly represents the desired learning features. This emphasizes the need to interrogate knowledge representation in biological LLMs.

Augmented Reality enables combining virtual data spaces with real-world environments through visual augmentations, transforming everyday environments into user interfaces of arbitrary type, size, and content. In the past, the development of Augmented Reality was mainly technology-driven. This made head-mounted Mixed Reality devices more common in research, industrial, or personal use cases. However, such devices are always human-centered, making it increasingly important to closely investigate and understand human factors within such applications and environments. Augmented Reality usage can reach from a simple information display to a dedicated device to present and analyze information visualizations. The growing data availability, amount, and complexity amplified the need and wish to generate insights through such visualizations. Those, in turn, can utilize human visual perception and Augmented Realitytextquoterights natural interactions, the potential to display three-dimensional data, or the stereoscopic display. In my thesis, I aim to deepen the understanding of how Augmented Reality applications must be designed to optimally adhere to human factors and ergonomics, especially in the area of visual data analysis. To address this challenge, I ground my thesis on three research questions: (1) How can we design such applications in a human-centered way? (2) What influence does the real-world environment have within such applications? (3) How can AR applications be combined with existing systems and devices? To answer those research questions, I explore different human properties and real-world environments that can affect the same environmenttextquoterights augmentations. For human factors, I investigate the competence in working with visualizations as visualization literacy, the visual perception of visualizations, and physical ergonomics like head movement. Regarding the environment, I examine two main factors: the visual backgroundtextquoterights influence on reading and working with immersive visualizations and the possibility of using alternative placement areas in Augmented Reality. Lastly, to explore future Augmented Reality systems, I designed and implemented Hybrid User Interfaces and authoring tools for immersive environments. Throughout the different projects, I used empirical, qualitative, and iterative methods in studying and designing immersive visualizations and applications. With that, I contribute to understanding how developers can apply human and environmental parameters for designing and creating future AR applications, especially for visual data analysis.

In multi-objective optimization, set-based quality indicators are a cornerstone of benchmarking and performance assessment. They capture the quality of a set of trade-off solutions by reducing it to a scalar number. One of the most commonly used set-based metrics is the R2 indicator, which describes the expected utility of a solution set to a decision-maker under a distribution of utility functions. Typically, this indicator is applied by discretizing this distribution of utility functions, yielding a weakly Pareto-compliant indicator. In consequence, adding a nondominated or dominating solution to a solution set may – but does not have to – improve the indicatortextquoterights value. In this paper, we reinvestigate the R2 indicator under the premise that we have a continuous, uniform distribution of (Tchebycheff) utility functions. We analyze its properties in detail, demonstrating that this continuous variant is indeed Pareto-compliant – that is, any beneficial solution will improve the metrictextquoterights value. Additionally, we provide an efficient computational procedure to compute this metric for bi-objective problems in O(NlogN). As a result, this work contributes to the state-of-the-art Pareto-compliant unary performance metrics, such as the hypervolume indicator, offering an efficient and promising alternative.

Cross-encoders distilled from large language models (LLMs) are often more effective re-rankers than cross-encoders fine-tuned on manually labeled data. However, the distilled models usually do not reach their teacher LLM's effectiveness. To investigate whether best practices for fine-tuning cross-encoders on manually labeled data (e.g., hard-negative sampling, deep sampling, and listwise loss functions) can help to improve LLM ranker distillation, we construct and release a new distillation dataset: Rank-DistiLLM. In our experiments, cross-encoders trained on Rank-DistiLLM reach the effectiveness of LLMs while being orders of magnitude more efficient. Our code and data is available at https://github.com/webis-de/msmarco-llm-distillation.

Active learning is an iterative labeling process that is used to obtain a small labeled subset, despite the absence of labeled data, thereby enabling to train a model for supervised tasks such as text classification.While active learning has made considerable progress in recent years due to improvements provided by pre-trained language models, there is untapped potential in the often neglected unlabeled portion of the data, although it is available in considerably larger quantities than the usually small set of labeled data. In this work, we investigate how self-training, a semi-supervised approach that uses a model to obtain pseudo-labels for unlabeled data, can be used to improve the efficiency of active learning for text classification. Building on a comprehensive reproduction of four previous self-training approaches, some of which are evaluated for the first time in the context of active learning or natural language processing, we introduce HAST, a new and effective self-training strategy, which is evaluated on four text classification benchmarks. Our results show that it outperforms the reproduced self-training approaches and reaches classification results comparable to previous experiments for three out of four datasets, using as little as 25% of the data. The code is publicly available at https://github.com/chschroeder/self-training-for-sample-efficient-active-learning.

Oversampling is commonly used to improve classifier performance for small tabular imbalanced datasets. State-of-the-art linear interpolation approaches can be used to generate synthetic samples from the convex space of the minority class. Generative networks are common deep learning approaches for synthetic sample generation. However, their scope on synthetic tabular data generation in the context of imbalanced classification is not adequately explored. In this article, we show that existing deep generative models perform poorly compared to linear interpolation-based approaches for imbalanced classification problems on small tabular datasets. To overcome this, we propose a deep generative model, ConvGeN that combines the idea of convex space learning with deep generative models. ConvGeN learns coefficients for the convex combinations of the minority class samples, such that the synthetic data is distinct enough from the majority class. Our benchmarking experiments demonstrate that our proposed model ConvGeN improves imbalanced classification on such small datasets, as compared to existing deep generative models, while being on par with the existing linear interpolation approaches. Moreover, we discuss how our model can be used for synthetic tabular data generation in general, even outside the scope of data imbalance, and thus improves the overall applicability of convex space learning.

The functionality of photoreceptors, rods, and cones is highly dependent on their outer segments (POS), a cellular compartment containing highly organized membranous structures that generate biochemical signals from incident light. While POS formation and degeneration are qualitatively assessed on microscopy images, reliable methodology for quantitative analyses is still limited. Here, we developed methods to quantify POS (QuaPOS) maturation and quality on retinal sections using automated image analyses. POS formation was examined during the development and in adulthood of wild-type mice via light microscopy (LM) and transmission electron microscopy (TEM). To quantify the number, size, shape, and fluorescence intensity of POS, retinal cryosections were immunostained for the cone POS marker S-opsin. Fluorescence images were used to train the robust classifier QuaPOS-LM based on supervised machine learning for automated image segmentation. Characteristic features of segmentation results were extracted to quantify the maturation of cone POS. Subsequently, this quantification method was applied to characterize POS degeneration in "cone photoreceptor function loss 1" mice. TEM images were used to establish the ultrastructural quantification method QuaPOS-TEM for the alignment of POS membranes. Images were analyzed using a custom-written MATLAB code to extract the orientation of membranes from the image gradient and their alignment (coherency). This analysis was used to quantify the POS morphology of wild-type and two inherited retinal degeneration ("retinal degeneration 19" and "rhodopsin knock-out") mouse lines. Both automated analysis technologies provided robust characterization and quantification of POS based on LM or TEM images. Automated image segmentation by the classifier QuaPOS-LM and analysis of the orientation of membrane stacks by QuaPOS-TEM using fluorescent or TEM images allowed quantitative evaluation of POS formation and quality. The assessments showed an increase in POS number, volume, and membrane coherency during wild-type postnatal development, while a decrease in all three observables was detected in different retinal degeneration mouse models. All the code used for the presented analysis is open source, including example datasets to reproduce the findings. Hence, the QuaPOS quantification methods are useful for in-depth characterization of POS on retinal sections in developmental studies, for disease modeling, or after therapeutic interventions affecting photoreceptors.

We consider the standard neural field equation with an exponential temporal kernel. We analyze the time-independent (static) and time-dependent (dynamic) bifurcations of the equilibrium solution and the emerging spatiotemporal wave patterns. We show that an exponential temporal kernel does not allow static bifurcations such as saddle-node, pitchfork, and in particular, static Turing bifurcations. However, the exponential temporal kernel possesses the important property that it takes into account the finite memory of past activities of neurons, which Green's function does not. Through a dynamic bifurcation analysis, we give explicit bifurcation conditions. Hopf bifurcations lead to temporally non-constant, but spatially constant solutions, but Turing--Hopf bifurcations generate spatially and temporally non-constant solutions, in particular, traveling waves. Bifurcation parameters are the coefficient of the exponential temporal kernel, the transmission speed of neural signals, the time delay rate of synapses, and the ratio of excitatory to inhibitory synaptic weights.

Machine learning has transitioned from an individ- ualistic approach to a collaborative one, enabling the collective effort to address increasingly complex challenges as they arise. One challenge that emerges is the management of a collaborative development process in machine learning projects. This paper outlines a collaborative environment KEENsight that leverages the benefits of a collaborative approach by orchestrating various open source tools. It establishes an optimal setting for code collaboration, model generation, data sharing, and the utilization of computational resources not limited to a single location. Through the integrating of these tools, KEENsight aims to streamline the development process and enhance productivity in machine learning.

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 442 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit ``breakthrough'' behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting.

Thrombocytopenia is a common side effect of cytotoxic chemotherapies, which is often dose-limiting. Predicting an individual's risk is of high clinical importance, as otherwise, a small subgroup of patients limits dosages for the overall population for safety reasons.

A high prevalence of mental health disorders among students combined with a lack of sufficient psychological counseling and treatment options makes new approaches to mental health support necessary. One possibility to strengthen the resilience and mental well-being of students in a scalable, low-threshold approach is the use of chatbots. While numerous work exists on chatbot-based cognitive behavioral therapy (CBT), the acceptance and usability of a resource-oriented approach to chatbot-based CBT has thus far not been investigated. Moreover, existing chatbot-based systems use a categorical model for the tracking of emotional states, thereby limiting the complexity of tracked emotions. This paper presents a chatbot-based approach for the tracking of dimensional emotional states and activities in the scope of resource-oriented CBT support for students. The developed concept and system are presented and investigated in a single-arm user study focusing on the usability and acceptance of the system with the primary target group (N=9). Results indicate excellent usability and good acceptance with some room for improvement for future iterations of the system.

Access to digital maps and geospatial information has become an integral part of modern applications to support mobility. However, for people with visual and mobility impairment, provided digital solutions are insufficient, target only one of the mentioned user groups, or are exclusively developed for outdoor environments. Research on accessible web-based indoor maps for people with impairments is still sparse. This paper presents the concept and prototype of an accessible web application that provides indoor map data and information on accessibility features of buildings to users with visual impairment, blindness, mobility impairment, and users without impairments. Results of a user study with 24 participants from all target groups show good to excellent usability (mean SUS score: 80), with room for improvement for users with severe visual impairment and blindness (mean SUS score: 66.7).

This paper presents TL;DR Progress, a new tool for exploring the literature on neural text summarization. It organizes 514textasciitildepapers based on a comprehensive annotation scheme for text summarization approaches and enables fine-grained, faceted search. Each paper was manually annotated to capture aspects such as evaluation metrics, quality dimensions, learning paradigms, challenges addressed, datasets, and document domains. In addition, a succinct indicative summary is provided for each paper, describing contextual factors, issues, and proposed solutions. The tool is available at urlhttps://www.tldr-progress.de, a demo video at urlhttps://youtu.be/uCVRGFvXUj8

Discriminative learning effectively predicts true object class for image classification. However, it often results in false positives for outliers, posing critical concerns in applications like autonomous driving and video surveillance systems. Previous attempts to address this challenge involved training image classifiers through contrastive learning using actual outlier data or synthesizing outliers for self-supervised learning. Furthermore, unsupervised generative modeling of inliers in pixel space has shown limited success for outlier detection. In this work, we introduce a quantile-based maximum likelihood objective for learning the inlier distribution to improve the outlier separation during inference. Our approach fits a normalizing flow to pre-trained discriminative features and detects the outliers according to the evaluated log-likelihood. The experimental evaluation demonstrates the effectiveness of our method as it surpasses the performance of the state-of-the-art unsupervised methods for outlier detection. The results are also competitive compared with a recent self-supervised approach for outlier detection. Our work allows to reduce dependency on well-sampled negative training data, which is especially important for domains like medical diagnostics or remote sensing.

State-of-the-art microscopes, as used in cell biology, are not only capable of capturing 3D images, but also permit manipulation of (sub-)cellular structures using techniques such as optical traps, optogenetics or laser ablation. However, such microscopes are still controlled using 2D interfaces, prohibiting actual 3-dimensional manipulation.We present microscenery, a virtual reality (VR) microscope control software, designed to facilitate 3D laser ablation experiments. We combine microscopy automation with VR rendering and intuitive controller-based input to empower biologists with the precision of laser-based techniques while providing the full 3D spatial context of their sample.We describe the design goals and architecture of the software and illustrate the potential of the system by conducting a brief expert review study for 3D ablation experiments. Our results suggest VR is not only an effective interface for microscopic manipulations, but can enable novel experiments which are either impossible with traditional 2D interfaces, or prohibitively time-consuming.

Statistical learning is the ability to extract and retain statistical regularities from the environment. In language, extracting statistical regularitiestextemdashso-called transitional probabilities, TPstextemdashis crucial for segmenting speech and learning new words. To investigate whether neural activity synchronizes with these statistical patterns, so-called neural frequency-tagging paradigms have been employed. Here, brain activity is recorded while a stream of syllables is presented at one fixed rate (e.g., 3.3 Hz) and rhythmic patterns of low-TP events are present at a different rate (e.g., 1.1 Hz). Such studies report neural tracking of both syllables and TPs; that is, for both acoustic and abstract events, respectively. However, recent criticism pointed out that the possible confounding of acoustic and abstract rhythms limits the interpretability of this work. Here, we present a new Python toolbox that minimizes this confounding by creating Artificial languages with Rhythmicity Controls (ARC). ARC features a phonological rhythmicity index (PRI) to quantify the confounding periodicity of phonological features at the TP rate. It implements a pseudo-random-walk approach to ensure unconfounded TP stationarity throughout streams of arbitrary length. We provide a step-by-step guide to generate artificial pseudowords based on a pre-specified syllable inventory and create controlled lexicon streams. Simulation results show that ARC lexicons minimize PRI and TP variance, not only relative to sets of random syllable groups, but also to benchmarking sets of artificial languages used in the prior literature. ARC should be used by future work in order to suppress the confounding of acoustics and TPs in artificial language learning experiments.Competing Interest StatementThe authors have declared no competing interest.

Given the growing popularity of graphs, data models like Property Graph and corresponding query languages like Cypher, have become popular in industry and research. However, real-time graph data analysis and management are increasingly important for today's businesses. Still, graph query languages lack the features to handle streaming graph data and their continuous query evaluation. % This demonstration showcases Seraph, a Cypher-based language supporting native continuous querying features. Seraph is formally defined, declarative, and backwards compatible with Cypher. This demonstration gives a comprehensive overview of the language with three use cases and a carefully design Web UI, allowing the execution of continuous queries on the property graphs stream.

Chimeric antigen receptor (CAR) T cell therapy has demonstrated remarkable efficacy in haematological malignancies. However, although viral vector-based methods are effective for CAR T cell manufacturing, they present significant biological, logistical and financial challenges. Non-viral vectors --- such as mRNA, transposon and precision genome editors --- offer a promising alternative, addressing issues of scalability and cost. mRNA and precision genome-editing technologies hold promise in facilitating the implementation of safe in vivo gene delivery for CAR T cell therapy. In this Primer, we discuss recent advancements in non-viral vectors for CAR immunotherapy, focusing on methodologies, clinical results, opportunities and limitations. We discuss permanent and transient gene engineering technologies, highlighting their ability to efficiently deliver genetic material and perform multiple genetic edits in target effector cells, such as T lymphocytes. We briefly describe delivery methods and preclinical development of ex vivo and in vivo T cell engineering using non-viral vectors. Additionally, we outline key considerations for clinical translation, such as reproducibility and good manufacturing practice. Through a comprehensive review of current literature and clinical trials, we underscore the potential of non-viral vectors to further optimize CAR immunotherapies, paving the way for more advanced and accessible cell-based treatments.