RECE: Reduced Cross-Entropy Loss for Large-Catalogue Sequential Recommenders
This addresses scalability challenges for practitioners deploying large-catalogue sequential recommenders, though it is incremental as it builds on existing cross-entropy loss methods.
The paper tackled the problem of excessive GPU memory consumption in sequential recommender systems with large item catalogs by introducing the RECE loss, which reduces training peak memory usage by up to 12 times while maintaining or exceeding the performance of full cross-entropy loss.
Scalability is a major challenge in modern recommender systems. In sequential recommendations, full Cross-Entropy (CE) loss achieves state-of-the-art recommendation quality but consumes excessive GPU memory with large item catalogs, limiting its practicality. Using a GPU-efficient locality-sensitive hashing-like algorithm for approximating large tensor of logits, this paper introduces a novel RECE (REduced Cross-Entropy) loss. RECE significantly reduces memory consumption while allowing one to enjoy the state-of-the-art performance of full CE loss. Experimental results on various datasets show that RECE cuts training peak memory usage by up to 12 times compared to existing methods while retaining or exceeding performance metrics of CE loss. The approach also opens up new possibilities for large-scale applications in other domains.