用强化学习动态筛选多向量嵌入，大幅压缩存储同时保持检索性能。
ReinPool: Reinforcement Learning Pooling Multi-Vector Embeddings for Retrieval System
- 通过强化学习自动挑选最具区分性的向量进行聚合
- 压缩比例达746至1249倍，保留原性能的76%~81%
- 适合需要高效检索系统的视觉语言模型应用
多向量嵌入模型通过标记级表示保留了细粒度的视觉与文本信息，但其表达力代价高昂：每个标记的嵌入存储使索引规模比单向量方法扩大1000倍以上，严重限制可扩展性。我们提出ReinPool，一种基于强化学习的框架，能够动态过滤并聚合多向量嵌入为紧凑且优化检索的表示。通过逆向检索目标和基于NDCG的奖励训练，ReinPool无需人工重要性标注即可识别并保留最具判别力的向量。在Vidore V2基准上，对三种视觉-语言嵌入模型测试显示，ReinPool将多向量表示压缩746至1249倍为单向量，同时恢复76%~81%的完整多向量检索性能。相比静态均值池化基线，ReinPool在NDCG@3上提升22%~33%，证明学习选择显著优于启发式聚合。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。