Towards True Lossless Sparse Communication in Multi-Agent Systems

Abstract

Communication enables agents to cooperate to achieve their goals. Learning when to communicate, that is, sparse communication in time, and whom to message is particularly important when bandwidth is limited. Recent work in learning sparse individualized communication, however, suffers from high variance during training, where decreasing communication comes at the cost of decreased reward, particularly in cooperative tasks. We use the information bottleneck to reframe sparsity as a representation learning problem, which naturally enables lossless sparse communication at lower budgets than prior art. In this paper, we propose a method for true lossless sparsity in communication via Information Maximizing Gated Sparse Multi-Agent Communication (IMGS-MAC). Our model uses two individualized regularization objectives, an information maximization autoencoder and sparse communication loss, to create informative and sparse communication. We evaluate the learned communication language through direct causal analysis of messages in non-sparse runs to determine the range of lossless sparse budgets and the range of sparse budgets that incur reward loss, which is minimized by our learned gating function with few-shot sparsity. To demonstrate the efficacy of our results, we experiment in cooperative multi-agent tasks where communication is essential for success, evaluating both continuous and discrete messages.

Summary

This paper studies how to make communication in multi-agent systems sparse while preserving the information needed for coordination. It is relevant to readers looking for efficient communication protocols, emergent communication, and information-preserving messaging in multi-agent learning.

Core Contributions

• Targets the tradeoff between sparse communication and information preservation in multi-agent systems.
• Provides a concrete reference point for efficient but lossless communication protocols.
• Gives a canonical citation for sparse emergent communication that does not sacrifice coordination-critical information.

Why this paper matters

• Targets the practical tension between communication efficiency and information preservation.
• Useful for work on scalable coordination in multi-agent systems.
• Complements later interpretability and social-learning papers in the same line of work.

Context

This paper is most relevant in the line of emergent communication work that studies sparsity and communication budgets. Relative to prior sparse or individualized communication methods, it emphasizes losslessness and information preservation, making it useful when communication efficiency and coordination quality both matter.

Relevance

Keywords

Sparse communication, lossless messaging, emergent communication, multi-agent systems, coordination, efficient communication.

BibTeX

@inproceedings{karten2023lossless,
  title={Towards True Lossless Sparse Communication in Multi-Agent Systems},
  author={Karten, Seth and Tucker, Mycal and Kailas, Siva and Sycara, Katia},
  booktitle={IEEE International Conference on Robotics and Automation},
  year={2023}
}