Towards Deep Symbolic Reinforcement Learning

TL;DR

This paper proposes a hybrid neural-symbolic reinforcement learning architecture that aims to address the limitations of deep reinforcement learning by enabling better reasoning, transfer learning, and interpretability, demonstrated through a simple video game prototype.

Contribution

It introduces an end-to-end neural-symbolic RL architecture combining neural networks with symbolic reasoning, showing promising preliminary results on a video game task.

Findings

The hybrid system learns effectively on a simple game.

It acquires symbolic rules that are human-readable.

It outperforms conventional DRL on a stochastic game variant.

Abstract

Deep reinforcement learning (DRL) brings the power of deep neural networks to bear on the generic task of trial-and-error learning, and its effectiveness has been convincingly demonstrated on tasks such as Atari video games and the game of Go. However, contemporary DRL systems inherit a number of shortcomings from the current generation of deep learning techniques. For example, they require very large datasets to work effectively, entailing that they are slow to learn even when such datasets are available. Moreover, they lack the ability to reason on an abstract level, which makes it difficult to implement high-level cognitive functions such as transfer learning, analogical reasoning, and hypothesis-based reasoning. Finally, their operation is largely opaque to humans, rendering them unsuitable for domains in which verifiability is important. In this paper, we propose an end-to-end reinforcement learning architecture comprising a neural back end and a symbolic front end with the potential to overcome each of these shortcomings. As proof-of-concept, we present a preliminary implementation of the architecture and apply it to several variants of a simple video game. We show that the resulting system -- though just a prototype -- learns effectively, and, by acquiring a set of symbolic rules that are easily comprehensible to humans, dramatically outperforms a conventional, fully neural DRL system on a stochastic variant of the game.