Towards Learning Abstractions via Reinforcement Learning

TL;DR

This paper introduces a neuro-symbolic reinforcement learning approach that enables multi-agent systems to develop higher-level communication abstractions, leading to faster convergence in collaborative tasks.

Contribution

It presents a novel method combining symbolic and machine learning techniques to extend communication primitives dynamically in multi-agent reinforcement learning.

Findings

Agents converge faster on collaborative tasks.

The approach enables development of higher-level communication concepts.

Shorter messages improve communication efficiency.

Abstract

In this paper we take the first steps in studying a new approach to synthesis of efficient communication schemes in multi-agent systems, trained via reinforcement learning. We combine symbolic methods with machine learning, in what is referred to as a neuro-symbolic system. The agents are not restricted to only use initial primitives: reinforcement learning is interleaved with steps to extend the current language with novel higher-level concepts, allowing generalisation and more informative communication via shorter messages. We demonstrate that this approach allow agents to converge more quickly on a small collaborative construction task.