EPR-Net: Constructing non-equilibrium potential landscape via a variational force projection formulation

We present a novel yet simple deep learning approach, dubbed EPR-Net, for constructing the potential landscape of high-dimensional non-equilibrium steady state (NESS) systems. The key idea of our approach is to utilize the fact that the negative potential gradient is the orthogonal projection of the driving force in a weighted Hilbert space with respect to the steady-state distribution. The constructed loss function also coincides with the entropy production rate (EPR) formula in NESS theory. This approach can be extended to dealing with dimensionality reduction and state-dependent diffusion coefficients in a unified fashion. The robustness and effectiveness of the proposed approach are demonstrated by numerical studies of several high-dimensional biophysical models with multi-stability, limit cycle, or strange attractor with non-vanishing noise.

Intrinsically Motivated Self-supervised Learning in Reinforcement Learning

In vision-based reinforcement learning (RL) tasks, it is prevalent to assign the auxiliary task with a surrogate self-supervised loss so as to obtain more semantic representations and improve sample efficiency. However, abundant information in self-supervised auxiliary tasks has been disregarded, since the representation learning part and the decision-making part are separated. To sufficiently utilize information in the auxiliary task, we present a simple yet effective idea to employ self-supervised loss as an intrinsic reward, called Intrinsically Motivated Self-Supervised learning in Reinforcement learning (IM-SSR). We formally show that the self-supervised loss can be decomposed as exploration for novel states and robustness improvement from nuisance elimination. IM-SSR can be effortlessly plugged into any reinforcement learning with self-supervised auxiliary objectives with nearly no additional cost. Combined with IM-SSR, the previous underlying algorithms achieve salient improvements on both sample efficiency and generalization in various vision-based robotics tasks from the DeepMind Control Suite, especially when the reward signal is sparse.