Invited : Kinematic from imaging: a lesson from using machine learning on active nematics

The primary probe for quantitative measurement of properties in active fluids is imaging. Accelerating the pipeline of data to physics and closing the loop between the learnt physics and developing data driven techniques to control the emergent behaviors of active fluids is critical for the development of active matter. In this work, I describe some work on 2D active nematics composed on microtubules and motor proteins that seeks to : a) learn kinematic information from imaging data, b) learn material parameters and dynamical equations from kinematic data and c) use control theory and reinforcement learning to identify experimentally realizable protocols for controlling the structure properties of emergent dynamical states of active fluids. While the specific ML implementations are customized to the physics of this particular active fluid, the research design and data science implications are generically applicable to diverse active matter systems extant in the literature.