CORE：通过重排序器蒸馏提升MLLM嵌入中的组合推理能力

摘要

基于MLLM的嵌入模型在组合检索方面仍然能力有限，通常无法区分包含相同概念但属性-物体绑定关系不同的场景。然而，同一骨干网络在被用作交叉注意力重排序器时却能够解决此类区分问题，这促使我们将重排序器的组合判断能力蒸馏到嵌入模型中。我们提出CORE，它综合生成跨越五个组合匹配级别的候选列表，并引入Rank-KL目标，训练嵌入模型复现重排序器的细粒度排序。我们进一步引入分级评估协议，并在相同数据和调参预算下比较对比学习、成对CoSENT与列表式Rank-KL。比较结果显示，CoSENT和Rank-KL对多级监督的利用均比对比学习更有效，其中Rank-KL取得了最强的整体性能。在三个组合推理基准（COLA、SUGARCREPE++、NEGBENCH）上，CORE-RERANKER-8B实现了82.7%的总平均，比Jina-Reranker高出10.7个百分点；而CORE-EMBED-8B在所有被评估的嵌入模型中取得了最佳总平均（0.666）。这些改进可迁移至MCMR基准，同时不牺牲在COCO和Flickr30K上的检索性能。

English

MLLM-based embedding models remain limited in compositional retrieval, often failing to distinguish scenes containing the same concepts but different attribute-object bindings. Yet the same backbone can resolve such distinctions when used as a cross-attentive reranker, motivating us to distill its compositional judgments into the embedding model. We propose CORE, which synthesizes candidate lists spanning five compositional matching levels and introduces a Rank-KL objective that trains the embedding model to reproduce the reranker's fine-grained ranking. We further introduce a graded evaluation protocol and compare contrastive learning, pairwise CoSENT, and listwise Rank-KL under the same data and tuning budget. Our comparison shows that both CoSENT and Rank-KL use the multi-level supervision more effectively than contrastive learning, with Rank-KL achieving the strongest overall performance. Across three compositional reasoning benchmarks (COLA, SUGARCREPE++, NEGBENCH), CORE-RERANKER-8B achieves an 82.7% total average, outperforming Jina-Reranker by 10.7 points, while CORE-EMBED-8B achieves the best total average (0.666) among all evaluated embedding models. The improvements transfer to the MCMR benchmark without sacrificing retrieval performance on COCO and Flickr30K.