Measuring the body to understand the mind


February 23, 2021 - 12:00pm - 1:00pm
About the Speaker
Jesse Marshall

Animal behavior is flexible: it can be adapted to meet the demands of new circumstances in the environment. But despite the recent revolution in machine learning, artificial motor systems lack this key ingredient, hindering their utility in the real world. Deciphering flexibility has been beyond our reach, in large part because it has been impossible to precisely measure animal movement across time and space. To address this, we developed CAPTURE, which uses 3D motion capture to achieve millisecond timescale and millimeter precision across dozens of body parts and the full repertoire of rat behavior, and DANNCE, which extends this capability across species and environments using geometric deep learning. CAPTURE and DANNCE enable novel quantitative inquiries into the function, mechanism, ontogeny, and evolution of behavior.  We then recorded naturalistic 3D kinematics in concert with continuous electrophysiological recordings of the striatum, a movement control center. Striatal neurons showed clear encoding of movement and behavior, but this encoding scheme could be reconfigured to meet new behavioral demands, for instance learned tasks. Similar encoding schema were seen in virtual rodents trained to imitate the behavior of real rodents. This integration of next-generation behavioral tracking, neurophysiology and deep imitation learning opens a pathway to identify principles of flexible motor control and translate them to artificial systems.