Compositional Obverter Communication Learning From Raw Visual Input

TL;DR

This paper demonstrates that neural agents can develop a compositional language directly from raw image pixels through an image description game, using the obverter technique, without relying on pre-engineered features.

Contribution

It introduces a method for training neural agents to learn compositional communication directly from raw visual input, mimicking human language development more closely.

Findings

Agents develop a compositional language from raw pixels

The language exhibits properties similar to human language

Zero-shot generalization demonstrates understanding of new concepts

Abstract

One of the distinguishing aspects of human language is its compositionality, which allows us to describe complex environments with limited vocabulary. Previously, it has been shown that neural network agents can learn to communicate in a highly structured, possibly compositional language based on disentangled input (e.g. hand- engineered features). Humans, however, do not learn to communicate based on well-summarized features. In this work, we train neural agents to simultaneously develop visual perception from raw image pixels, and learn to communicate with a sequence of discrete symbols. The agents play an image description game where the image contains factors such as colors and shapes. We train the agents using the obverter technique where an agent introspects to generate messages that maximize its own understanding. Through qualitative analysis, visualization and a zero-shot test, we show that the agents can develop, out of raw image pixels, a language with compositional properties, given a proper pressure from the environment.