Reward functions in RL turn out to be surprisingly similar to vector fields—they can be conservative gradient fields and you can take their divergence and curl. These operations are related to RL concepts such as optimal policies. See our new paper arxiv.org/abs/2208.09570. 🧵👇 A well-known property of reward functions is that you can add a potential shaping term to them without changing the optimal policy (under any transition dynamics). This shaping term is induced by an arbitrary function on the state space, the "potential" Φ.

Excited to announce that my paper with @maurice_weiler on Steerable Partial Differential Operators has been accepted to #iclr2022! Steerable PDOs bring equivariance to differential operators. Preprint: arxiv.org/abs/2106.10163 (1/N) Equivariance has become a popular topic in deep learning, but it has played a huge role in physics long before that. So wouldn't it be great if we could bring equivariant deep learning and physics closer together, to transfer more ideas? (2/N)