Inductive Bias for Emergent Communication in a Continuous Setting

TL;DR

This paper investigates how an inductive bias influences the emergence of communication protocols in multi-agent reinforcement learning, particularly focusing on continuous messages and their comparison with discrete communication.

Contribution

It introduces an inductive bias to facilitate the development of effective communication protocols in continuous message spaces within multi-agent RL environments.

Findings

Inductive bias improves communication protocol quality in toy environments.

The bias benefits both continuous and discrete message communication.

Combining inductive bias with reinforcement learning enhances emergent communication.

Abstract

We study emergent communication in a multi-agent reinforcement learning setting, where the agents solve cooperative tasks and have access to a communication channel. The communication channel may consist of either discrete symbols or continuous variables. We introduce an inductive bias to aid with the emergence of good communication protocols for continuous messages, and we look at the effect this type of inductive bias has for continuous and discrete messages in itself or when used in combination with reinforcement learning. We demonstrate that this type of inductive bias has a beneficial effect on the communication protocols learnt in two toy environments, Negotiation and Sequence Guess.