1:00 pm
GCIS and Zoom Room W301
"Machine learning for Bio-Active Matter"
Continuum theories effectively describe many-body systems out of equilibrium in terms of a few macroscopic parameters. Such parameters are difficult to determine from microscopics, particularly in active and biological matter where they are proxies for energy-injecting components. Here, we demonstrate how machine learning can help characterize systems which resist traditional continuum modeling. We use active nematics to demonstrate that neural networks can determine how multiple hydrodynamic parameters change as a function of space and time due to engineered microscopic behavior in experiments. We can also forecast the evolution of these chaotic systems solely from image-sequences of their past, without requiring any knowledge of their underlying physics. We extend our physics-agnostic approach to predict force generation in cells directly from fluorescently-labeled protein distributions. Beyond achieving excellent predictive accuracy, our neural networks also serve as a stepping stone to automatically construct effective continuum equations for cell mechanics. Our study presents neural networks as an integral tool for characterizing diverse active and biological systems, even when no detailed knowledge of their underlying mechanisms exists. Host: Sure Vaikuntanathan, svaikunt@uchicago.edu or by phone at 773-702-7256. Persons who may need assistance please contact Brenda Thomas at bthomas@uchicago.edu or by phone at 773-702-7156.