Col-Bandit：面向延迟交互检索的查询时Top-$K$估计
Col-Bandit: Query-Time Top-$K$ Estimation for Late-Interaction Retrieval
AI总结：
提出Col-Bandit方法，通过批量揭示矩阵条目并维护置信区间，在查询时高效估计多向量延迟交互检索中的Top-$K$结果，大幅减少计算量且保持高保真度。
摘要：
多向量延迟交互检索器（如ColBERT）实现了最先进的质量，但其查询时成本主要来自对每个候选文档穷举计算词级MaxSim交互。N个候选与T个查询词的MaxSim得分构成N×T矩阵，其行和为延迟交互得分，而识别top-$K$通常不需要所有条目。我们提出Col-Bandit，一种穷举MaxSim top-$K$的查询时估计方法：它批量揭示矩阵条目，维护每个候选得分的有限总体Bernstein-Serfling置信区间，并永久丢弃上界低于第K大下界的文档，仅计算分离top-$K$所需的单元格。一个松弛旋钮$α_{\mathrm{ef}}\in(0,1]$调节计算-保真度权衡。我们部署$α_{\mathrm{ef}}{=}0.2$，而$α_{\mathrm{ef}}{=}1$在简化半径下提供$δ$-PAC保证。在BEIR和REAL-MM-RAG上，Col-Bandit在每个语料库上对穷举top-$5$保持$\geq 90\%$的保真度，同时将MaxSim FLOPs减少高达${\sim}8\times$，在x86和ARM上实现高达${\sim}13\times$的单线程CPU加速。作为即插即用的重排序层，它无需重新训练或更改索引。
英文摘要：
Multi-vector late-interaction retrievers such as ColBERT achieve state-of-the-art quality, but their query-time cost is dominated by exhaustively computing token-level MaxSim interactions for every candidate document. The MaxSim scores of $N$ candidates against $T$ query tokens form an $N\times T$ matrix whose row-sums are the late-interaction scores, and identifying the top-$K$ rarely requires every entry. We introduce Col-Bandit, a query-time estimator of the exhaustive-MaxSim top-$K$: it reveals matrix entries in batches, maintains a finite-population Bernstein-Serfling confidence interval on each candidate's score, and permanently drops any document whose upper bound falls below the $K$-th largest lower bound, computing only the cells needed to separate the top-$K$. A single relaxation knob $α_{\mathrm{ef}}\in(0,1]$ tunes the compute-fidelity trade-off. We deploy $α_{\mathrm{ef}}{=}0.2$, while $α_{\mathrm{ef}}{=}1$ admits a $δ$-PAC guarantee under a simplified radius. On BEIR and REAL-MM-RAG, Col-Bandit preserves $\geq 90\%$ fidelity to the exhaustive top-$5$ on every corpus while cutting MaxSim FLOPs by up to ${\sim}8\times$, for up to ${\sim}13\times$ single-thread CPU speedups across x86 and ARM. A drop-in reranking layer, it needs no retraining or index changes.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
