Discrete Messages Improve Communication Efficiency among Isolated Intelligent Agents

TL;DR

The empirical findings indicate that, in scenarios where agents are exposed to different data, communicating through sentences composed of discrete tokens offers the best inter-agent communication efficiency.

Abstract

Individuals, despite having varied life experiences and learning processes, can communicate effectively through languages. This study aims to explore the efficiency of language as a communication medium. We put forth two specific hypotheses: First, discrete messages are more effective than continuous ones when agents have diverse personal experiences. Second, communications using multiple discrete tokens are more advantageous than those using a single token. To valdate these hypotheses, we designed multi-agent machine learning experiments to assess communication efficiency using various information transmission methods between speakers and listeners. Our empirical findings indicate that, in scenarios where agents are exposed to different data, communicating through sentences composed of discrete tokens offers the best inter-agent communication efficiency. The limitations of our finding include lack of systematic advantages over other more sophisticated encoder-decoder model such as variational autoencoder and lack of evluation on non-image dataset, which we will leave for future studies.

Figures (16)

• Figure 1: Left: Lewis Game. Right: Multi-token Discrete Mechanism. The communication vector is initially divided into multiple discretization tokens. Each token goes through separate discretization, where it is quantized to the nearest neighbor within a shared collection of latent codebook vectors. Subsequently, the discretization tokens are concatenated back together to form a vector with the same shape as the original one.
• Figure 2: Ten agents' understanding of the same language. Left: Feature distribution of latent codebooks for different agents. Right: Similarity of different latent codebooks. The understanding of this language is different for each agent.
• Figure 3: Training and validation of agents. Each agent has its own dataset during training. Upon completion of learning, one agent interacts with the other agent . This is our core methodology, where in this validation scenario, the advantage of discrete language in communication between agents is determined based on the reconstruction losses of information.
• Figure 4: Example images for the reconstruction task. Left: Original image. Right: Reconstructed image.
• Figure 5: Communication loss on three types of models. The x-axis represents different overlap ratios, and the y-axis represents the communication loss between agents.