Bridging the Gap: Unpacking the Hidden Challenges in Knowledge Distillation for Online Ranking Systems

TL;DR

This paper explores the unique challenges of applying knowledge distillation to online recommender systems, addressing data shifts, teacher configuration, and efficient label sharing, with practical evaluations on Google’s video recommendation systems.

Contribution

It introduces a robust KD framework tailored for recommender systems, tackling data distribution shifts, teacher configuration optimization, and rapid label sharing, validated through large-scale Google experiments.

Findings

Significant performance improvements in student models.

Effective mitigation of data distribution shifts.

Reliable and efficient teacher label generation.

Abstract

Knowledge Distillation (KD) is a powerful approach for compressing a large model into a smaller, more efficient model, particularly beneficial for latency-sensitive applications like recommender systems. However, current KD research predominantly focuses on Computer Vision (CV) and NLP tasks, overlooking unique data characteristics and challenges inherent to recommender systems. This paper addresses these overlooked challenges, specifically: (1) mitigating data distribution shifts between teacher and student models, (2) efficiently identifying optimal teacher configurations within time and budgetary constraints, and (3) enabling computationally efficient and rapid sharing of teacher labels to support multiple students. We present a robust KD system developed and rigorously evaluated on multiple large-scale personalized video recommendation systems within Google. Our live experiment results demonstrate significant improvements in student model performance while ensuring consistent and reliable generation of high quality teacher labels from a continuous data stream of data.