Self-Attentional Credit Assignment for Transfer in Reinforcement Learning

TL;DR

This paper introduces SECRET, a novel transfer learning method in reinforcement learning that uses self-attention for credit assignment, improving sample efficiency by transferring task invariants through reward modifications.

Contribution

It proposes SECRET, a self-attentional credit assignment approach that enables transfer in RL by learning to assign credit separately and modifying rewards, compatible with any RL algorithm.

Findings

SECRET improves transfer efficiency in RL tasks.

The method is compatible with various RL algorithms.

It effectively captures structural invariants for transfer.

Abstract

The ability to transfer knowledge to novel environments and tasks is a sensible desiderata for general learning agents. Despite the apparent promises, transfer in RL is still an open and little exploited research area. In this paper, we take a brand-new perspective about transfer: we suggest that the ability to assign credit unveils structural invariants in the tasks that can be transferred to make RL more sample-efficient. Our main contribution is SECRET, a novel approach to transfer learning for RL that uses a backward-view credit assignment mechanism based on a self-attentive architecture. Two aspects are key to its generality: it learns to assign credit as a separate offline supervised process and exclusively modifies the reward function. Consequently, it can be supplemented by transfer methods that do not modify the reward function and it can be plugged on top of any RL algorithm.