MICE-P：从仅重排序走向端到端检索
原标题：MICE to Meet You: From Re-ranking Only to End-to-End Retrieval
AI 摘要
该论文提出 MICE-P，首个基于 MICE 架构的端到端检索流水线，通过引入低维瓶颈将 MICE 的文档表示映射到与 PLAID 兼容的空间，从而实现细粒度文档表示的高效存储与索引。作者在匹配实验条件下用 Ettin 家族训练了 MICE、ColBERT 和 cross-encoder 的可比模型进行受控比较，结果显示 MICE-P 在保持 cross-encoder 效果的同时达到与 late-interaction 检索相当的吞吐量，定义了一类新的检索架构。
正文节选
MICE to Meet You: From Re-ranking Only to End-to-End Retrieval Abstract Late-interaction models offer state-of-the art efficiency-effectiveness trade-offs for neural retrieval by enabling document representations to be pre-computed and efficiently indexed. Yet, their relatively simple interaction mechanisms, typically MaxSim, can limit their effectiveness compared to cross-encoders. MICE, a recently introduced re-ranking architecture, aims to combine cross-encoder-like effectiveness through exp