Towards Learning Abstractions via Reinforcement Learning

2022/12/28 by Erik Jergéus, Jergéus, Erik, Leo Karlsson Oinonen +5
Computer Science · #Evolutionary Algorithms and Applications #Neural Networks and Applications

paper · pdf · doi:10.48550/arxiv.2212.13980

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.

Related

• PACE: Procedural Abstractions for Communicating Efficiently
• Emergence of Language with Multi-agent Games: Learning to Communicate with Sequences of Symbols
• Learning to Communicate with Deep Multi-Agent Reinforcement Learning
• Towards Language-Augmented Multi-Agent Deep Reinforcement Learning
• Abstraction for Deep Reinforcement Learning