Improving Training Stability for Multitask Ranking Models in Recommender Systems

TL;DR

This paper addresses the challenge of training instability in large, complex multitask ranking models for recommender systems, proposing a new algorithm that enhances stability without sacrificing convergence.

Contribution

It identifies properties causing instability, analyzes why existing solutions fail, and introduces a novel algorithm to improve training stability in real-world recommender models.

Findings

The proposed algorithm significantly improves training stability.

It maintains model convergence while enhancing stability.

Experiments on YouTube data validate effectiveness.

Abstract

Recommender systems play an important role in many content platforms. While most recommendation research is dedicated to designing better models to improve user experience, we found that research on stabilizing the training for such models is severely under-explored. As recommendation models become larger and more sophisticated, they are more susceptible to training instability issues, i.e., loss divergence, which can make the model unusable, waste significant resources and block model developments. In this paper, we share our findings and best practices we learned for improving the training stability of a real-world multitask ranking model for YouTube recommendations. We show some properties of the model that lead to unstable training and conjecture on the causes. Furthermore, based on our observations of training dynamics near the point of training instability, we hypothesize why existing solutions would fail, and propose a new algorithm to mitigate the limitations of existing solutions. Our experiments on YouTube production dataset show the proposed algorithm can significantly improve training stability while not compromising convergence, comparing with several commonly used baseline methods.