Multi-lingual agents through multi-headed neural networks

TL;DR

This paper introduces multi-headed neural networks for multi-lingual communication in multi-agent reinforcement learning, enabling agents to learn and maintain multiple languages simultaneously, thus mitigating catastrophic forgetting.

Contribution

The paper proposes a novel multi-headed neural network approach inspired by continual learning to enable agents to be multi-lingual in multi-agent reinforcement learning environments.

Findings

Agents can maintain multiple languages with the proposed method.

Existing approaches fail to sustain multiple languages effectively.

The approach is validated on a referential MNIST communication game.

Abstract

This paper considers cooperative Multi-Agent Reinforcement Learning, focusing on emergent communication in settings where multiple pairs of independent learners interact at varying frequencies. In this context, multiple distinct and incompatible languages can emerge. When an agent encounters a speaker of an alternative language, there is a requirement for a period of adaptation before they can efficiently converse. This adaptation results in the emergence of a new language and the forgetting of the previous language. In principle, this is an example of the Catastrophic Forgetting problem which can be mitigated by enabling the agents to learn and maintain multiple languages. We take inspiration from the Continual Learning literature and equip our agents with multi-headed neural networks which enable our agents to be multi-lingual. Our method is empirically validated within a referential MNIST based communication game and is shown to be able to maintain multiple languages where existing approaches cannot.