UniMixer: A Unified Architecture for Scaling Laws in Recommendation Systems

TL;DR

UniMixer introduces a unified architecture for recommendation models that unifies mainstream scaling blocks, optimizes token mixing, and enhances scaling efficiency through a lightweight module, validated by extensive experiments.

Contribution

The paper proposes UniMixer, a unified framework that unifies different recommendation model architectures and introduces a learnable token mixing module for improved scaling.

Findings

UniMixer achieves superior scaling performance in recommendation systems.

The UniMixing-Lite module reduces parameters and computational cost significantly.

Extensive experiments verify the effectiveness of UniMixer in both offline and online settings.

Abstract

In recent years, the scaling laws of recommendation models have attracted increasing attention, which govern the relationship between performance and parameters/FLOPs of recommenders. Currently, there are three mainstream architectures for achieving scaling in recommendation models, namely attention-based, TokenMixer-based, and factorization-machine-based methods, which exhibit fundamental differences in both design philosophy and architectural structure. In this paper, we propose a unified scaling architecture for recommendation systems, namely \textbf{UniMixer}, to improve scaling efficiency and establish a unified theoretical framework that unifies the mainstream scaling blocks. By transforming the rule-based TokenMixer to an equivalent parameterized structure, we construct a generalized parameterized feature mixing module that allows the token mixing patterns to be optimized and learned during model training. Meanwhile, the generalized parameterized token mixing removes the constraint in TokenMixer that requires the number of heads to be equal to the number of tokens. Furthermore, we establish a unified scaling module design framework for recommender systems, which bridges the connections among attention-based, TokenMixer-based, and factorization-machine-based methods. To further boost scaling ROI, a lightweight UniMixing module is designed, \textbf{UniMixing-Lite}, which further compresses the model parameters and computational cost while significantly improve the model performance. The scaling curves are shown in the following figure. Extensive offline and online experiments are conducted to verify the superior scaling abilities of \textbf{UniMixer}.