ReinPool: Reinforcement Learning Pooling Multi-Vector Embeddings for Retrieval System
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866911368503885824 |
|---|---|
| author | Cha, Sungguk Kim, DongWook Kim, Mintae Han, Youngsub Jeon, Byoung-Ki Lee, Sangyeob |
| author_facet | Cha, Sungguk Kim, DongWook Kim, Mintae Han, Youngsub Jeon, Byoung-Ki Lee, Sangyeob |
| contents | Multi-vector embedding models have emerged as a powerful paradigm for document retrieval, preserving fine-grained visual and textual details through token-level representations. However, this expressiveness comes at a staggering cost: storing embeddings for every token inflates index sizes by over $1000\times$ compared to single-vector approaches, severely limiting scalability. We introduce \textbf{ReinPool}, a reinforcement learning framework that learns to dynamically filter and pool multi-vector embeddings into compact, retrieval-optimized representations. By training with an inverse retrieval objective and NDCG-based rewards, ReinPool identifies and retains only the most discriminative vectors without requiring manual importance annotations. On the Vidore V2 benchmark across three vision-language embedding models, ReinPool compresses multi-vector representations by $746$--$1249\times$ into single vectors while recovering 76--81\% of full multi-vector retrieval performance. Compared to static mean pooling baselines, ReinPool achieves 22--33\% absolute NDCG@3 improvement, demonstrating that learned selection significantly outperforms heuristic aggregation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_07125 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | ReinPool: Reinforcement Learning Pooling Multi-Vector Embeddings for Retrieval System Cha, Sungguk Kim, DongWook Kim, Mintae Han, Youngsub Jeon, Byoung-Ki Lee, Sangyeob Information Retrieval Computation and Language Computer Vision and Pattern Recognition Multi-vector embedding models have emerged as a powerful paradigm for document retrieval, preserving fine-grained visual and textual details through token-level representations. However, this expressiveness comes at a staggering cost: storing embeddings for every token inflates index sizes by over $1000\times$ compared to single-vector approaches, severely limiting scalability. We introduce \textbf{ReinPool}, a reinforcement learning framework that learns to dynamically filter and pool multi-vector embeddings into compact, retrieval-optimized representations. By training with an inverse retrieval objective and NDCG-based rewards, ReinPool identifies and retains only the most discriminative vectors without requiring manual importance annotations. On the Vidore V2 benchmark across three vision-language embedding models, ReinPool compresses multi-vector representations by $746$--$1249\times$ into single vectors while recovering 76--81\% of full multi-vector retrieval performance. Compared to static mean pooling baselines, ReinPool achieves 22--33\% absolute NDCG@3 improvement, demonstrating that learned selection significantly outperforms heuristic aggregation. |
| title | ReinPool: Reinforcement Learning Pooling Multi-Vector Embeddings for Retrieval System |
| topic | Information Retrieval Computation and Language Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2601.07125 |