Fokker-Planck-Inverse Reinforcement Learning

Explore the innovative Fokker-Planck-Inverse Reinforcement Learning technique in this hour-long talk by Krishna Garikipati from the University of Michigan. Delve into the challenges of Inverse Reinforcement Learning (IRL) when transition dynamics are unknown, and discover how the Fokker-Planck equation from mean-field theory can be leveraged to infer these transitions. Learn about the proposed isomorphism between time-discrete Fokker-Planck and Markov decision processes, and how it forms the basis for a novel physics-aware IRL algorithm. Witness the application of this groundbreaking approach to both synthetic benchmarks and real-world biological problems in cancer cell dynamics. Gain insights from Garikipati's extensive background in computational science, biophysics, and data-driven approaches as he presents this collaborative research effort.

DDPS | Fokker-Planck-Inverse Reinforcement Learning