A Review of Sparse Expert Models in Deep Learning

TL;DR

Sparse expert models, a concept from thirty years ago, are now central in deep learning, enabling large, efficient models across multiple domains by acting on only subsets of parameters per example.

Contribution

This paper reviews the evolution, algorithms, and recent advances of sparse expert models, highlighting their significance and future research directions.

Findings

Sparse expert models enable large-scale, efficient deep learning.

They have achieved significant improvements in NLP, vision, and speech.

The paper identifies key areas for future research in sparse models.

Abstract

Sparse expert models are a thirty-year old concept re-emerging as a popular architecture in deep learning. This class of architecture encompasses Mixture-of-Experts, Switch Transformers, Routing Networks, BASE layers, and others, all with the unifying idea that each example is acted on by a subset of the parameters. By doing so, the degree of sparsity decouples the parameter count from the compute per example allowing for extremely large, but efficient models. The resulting models have demonstrated significant improvements across diverse domains such as natural language processing, computer vision, and speech recognition. We review the concept of sparse expert models, provide a basic description of the common algorithms, contextualize the advances in the deep learning era, and conclude by highlighting areas for future work.