Modular Networks: Learning to Decompose Neural Computation

TL;DR

This paper introduces a flexible training algorithm for modular neural networks that learns to decompose computation end-to-end, enabling efficient scaling and interpretable specialization in image and language tasks.

Contribution

It presents a novel end-to-end training method for modular networks that does not require regularization for diversity, improving performance and interpretability.

Findings

Achieves superior performance on image recognition and language modeling tasks.

Modules specialize in interpretable contexts.

Training method does not rely on regularization for module diversity.

Abstract

Scaling model capacity has been vital in the success of deep learning. For a typical network, necessary compute resources and training time grow dramatically with model size. Conditional computation is a promising way to increase the number of parameters with a relatively small increase in resources. We propose a training algorithm that flexibly chooses neural modules based on the data to be processed. Both the decomposition and modules are learned end-to-end. In contrast to existing approaches, training does not rely on regularization to enforce diversity in module use. We apply modular networks both to image recognition and language modeling tasks, where we achieve superior performance compared to several baselines. Introspection reveals that modules specialize in interpretable contexts.