Investigating Biologically-Inspired Approaches for Continual Reinforcement Learning

Olya Mastikhina, Golnaz Mesbahi, Martha White

Keywords: reinforcement learning, continual reinforcement learning, energy constraints, indexing

Abstract:

Despite the brain's natural ability to continuously learn, biological insights are rarely leveraged in continual reinforcement learning (RL).
In this paper, we aim to help bridge this gap by briefly examining four under-investigated biologically-motivated modifications within the context of continual RL: energy minimization, wire length constraints, sparse distributed memory multilayer perceptrons, and fuzzy tiling activations. We show that some of these modifications help increase plasticity and decrease catastrophic forgetting, and we provide an analysis of the learned representations.

Submission Number: 34