Mixed Dimension Embeddings with Application to Memory-Efficient Recommendation Systems

TL;DR

This paper introduces mixed dimension embeddings that scale embedding size with query frequency, significantly reducing memory usage in recommendation systems while maintaining or improving accuracy.

Contribution

It proposes a novel mixed dimension embedding architecture with theoretical analysis and empirical validation demonstrating substantial memory savings and performance retention.

Findings

Memory usage reduced by up to 16X

Accuracy improved by 0.1% on Criteo dataset

Maintains performance with fewer parameters

Abstract

Embedding representations power machine intelligence in many applications, including recommendation systems, but they are space intensive -- potentially occupying hundreds of gigabytes in large-scale settings. To help manage this outsized memory consumption, we explore mixed dimension embeddings, an embedding layer architecture in which a particular embedding vector's dimension scales with its query frequency. Through theoretical analysis and systematic experiments, we demonstrate that using mixed dimensions can drastically reduce the memory usage, while maintaining and even improving the ML performance. Empirically, we show that the proposed mixed dimension layers improve accuracy by 0.1% using half as many parameters or maintain it using 16X fewer parameters for click-through rate prediction task on the Criteo Kaggle dataset.