dorsal/arxiv
View SchemaReinPool: Reinforcement Learning Pooling Multi-Vector Embeddings for Retrieval System
| Authors | Sungguk Cha, DongWook Kim, Mintae Kim, Youngsub Han, Byoung-Ki Jeon, Sangyeob Lee |
|---|---|
| Categories | |
| ArXiv ID | 2601.07125vv1 |
| URL | https://arxiv.org/abs/2601.07125 |
| License | http://creativecommons.org/licenses/by-nc-nd/4.0/ |
Abstract
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.
{
"annotation_id": "6d536a1c-7847-4819-aa3d-6befc9dfbe01",
"date_created": "2026-02-17T05:53:12.638000Z",
"date_modified": "2026-02-17T05:53:12.638000Z",
"file_hash": "a8317812a3b4d671bc1783e55b03f6eeb7a083f7be23eb18790c7a132c5634d8",
"private": false,
"record": {
"abstract": "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.",
"arxiv_id": "2601.07125",
"authors": [
"Sungguk Cha",
"DongWook Kim",
"Mintae Kim",
"Youngsub Han",
"Byoung-Ki Jeon",
"Sangyeob Lee"
],
"categories": [
"cs.IR",
"cs.CL",
"cs.CV"
],
"license": "http://creativecommons.org/licenses/by-nc-nd/4.0/",
"title": "ReinPool: Reinforcement Learning Pooling Multi-Vector Embeddings for Retrieval System",
"url": "https://arxiv.org/abs/2601.07125",
"version": "v1"
},
"schema_id": "dorsal/arxiv",
"source": {
"execution_id": "16cf0bf4-c67c-41c8-84f1-a5c32f953f37",
"id": "arXiv Dataset",
"type": "Model",
"variant": "snapshot-2026-01-17",
"version": "0.1.0"
},
"user_id": 1000002
}