Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2602.17050 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916013512065024 |
|---|---|
| author | Zhao, Ziliang Xue, Bi Lin, Emma Lu, Tianqi Zhou, Mengjiao Vartak, Kaustubh Ali-Zade, Shakhzod Li, Tao Kuang, Bin Jian, Rui Wen, Bin van der Staay, Dennis Bao, Yixin Li, Eddy Deng, Chao Wei, Henry Liu, Songbin Wang, Qifan Ren, Kai |
| author_facet | Zhao, Ziliang Xue, Bi Lin, Emma Lu, Tianqi Zhou, Mengjiao Vartak, Kaustubh Ali-Zade, Shakhzod Li, Tao Kuang, Bin Jian, Rui Wen, Bin van der Staay, Dennis Bao, Yixin Li, Eddy Deng, Chao Wei, Henry Liu, Songbin Wang, Qifan Ren, Kai |
| contents | Embedding tables are critical components of large-scale recommendation systems, facilitating the efficient mapping of high-cardinality categorical features into dense vector representations. However, as the volume of unique IDs expands, traditional hash-based indexing methods suffer from collisions that degrade model performance and personalization quality. We present Multi-Probe Zero Collision Hash (MPZCH), a novel indexing mechanism based on linear probing that effectively mitigates embedding collisions. With reasonable table sizing, it often eliminates these collisions entirely while maintaining production-scale efficiency. MPZCH utilizes auxiliary tensors and high-performance CUDA kernels to implement configurable probing and active eviction policies. By retiring obsolete IDs and resetting reassigned slots, MPZCH prevents the stale embedding inheritance typical of hash-based methods, ensuring new features learn effectively from scratch. Despite its collision-mitigation overhead, the system maintains training QPS and inference latency comparable to existing methods. Rigorous online experiments demonstrate that MPZCH achieves zero collisions for user embeddings and significantly improves item embedding freshness and quality. The solution has been released within the open-source TorchRec library for the broader community. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_17050 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Multi-Probe Zero Collision Hash (MPZCH): Mitigating Embedding Collisions and Enhancing Model Freshness in Large-Scale Recommenders Zhao, Ziliang Xue, Bi Lin, Emma Lu, Tianqi Zhou, Mengjiao Vartak, Kaustubh Ali-Zade, Shakhzod Li, Tao Kuang, Bin Jian, Rui Wen, Bin van der Staay, Dennis Bao, Yixin Li, Eddy Deng, Chao Wei, Henry Liu, Songbin Wang, Qifan Ren, Kai Machine Learning Embedding tables are critical components of large-scale recommendation systems, facilitating the efficient mapping of high-cardinality categorical features into dense vector representations. However, as the volume of unique IDs expands, traditional hash-based indexing methods suffer from collisions that degrade model performance and personalization quality. We present Multi-Probe Zero Collision Hash (MPZCH), a novel indexing mechanism based on linear probing that effectively mitigates embedding collisions. With reasonable table sizing, it often eliminates these collisions entirely while maintaining production-scale efficiency. MPZCH utilizes auxiliary tensors and high-performance CUDA kernels to implement configurable probing and active eviction policies. By retiring obsolete IDs and resetting reassigned slots, MPZCH prevents the stale embedding inheritance typical of hash-based methods, ensuring new features learn effectively from scratch. Despite its collision-mitigation overhead, the system maintains training QPS and inference latency comparable to existing methods. Rigorous online experiments demonstrate that MPZCH achieves zero collisions for user embeddings and significantly improves item embedding freshness and quality. The solution has been released within the open-source TorchRec library for the broader community. |
| title | Multi-Probe Zero Collision Hash (MPZCH): Mitigating Embedding Collisions and Enhancing Model Freshness in Large-Scale Recommenders |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2602.17050 |