Topic: Enabling Data-Driven Discovery in Biology: Statistical Learning of Interpretable Mathematical Models from Microscopy Videos
In the last decade, advancements in ML algorithms, computing power and high-resolution sensing technologies has enabled us to understand and decode complex physical systems through the eyes of “Big-data”. This has generated acute interest in using ML paradigms for studying complex dynamical systems. In my talk, I will review methods based on sparse-regression for learning interpretable mathematical models (ODE,PDE) of dynamical systems directly from their spatio-temporal measurements. The presentation includes details of formulation of the non-convex problem and its solution, statistical theory for bounding the # of false-positives, and the results from application to both simulation and experimental data-sets. Future work on encoding prior knowledge like symmetries and conservation laws, and the potential of deep-neural architectures for learning interpretable models will be outlined.