Hi, we publish new embedding models which perform well on fine-grained multimodal retrieval and compositional reasoning tasks.</p>\n","updatedAt":"2026-09-04T02:18:34.011Z","author":{"_id":"64dc29d9b5d625e0e9a6ecb9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/QxGBsnk1cNsBEPqSx4ae-.jpeg","fullname":"Tingyu Song","name":"songtingyu","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":4,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.917273223400116},"editors":["songtingyu"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/QxGBsnk1cNsBEPqSx4ae-.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2609.04083","authors":[{"_id":"6a9a278c8f7c3b7557239421","name":"Tingyu Song","hidden":false},{"_id":"6a9a278c8f7c3b7557239422","name":"Mingxin Li","hidden":false},{"_id":"6a9a278c8f7c3b7557239423","name":"Yanzhao Zhang","hidden":false},{"_id":"6a9a278c8f7c3b7557239424","name":"Dingkun Long","hidden":false},{"_id":"6a9a278c8f7c3b7557239425","name":"Chu Liu","hidden":false},{"_id":"6a9a278c8f7c3b7557239426","name":"Pengjun Xie","hidden":false},{"_id":"6a9a278c8f7c3b7557239427","name":"Yilun Zhao","hidden":false},{"_id":"6a9a278c8f7c3b7557239428","name":"Shu Wu","hidden":false}],"publishedAt":"2026-09-03T00:00:00.000Z","submittedOnDailyAt":"2026-09-04T00:00:00.000Z","title":"CORE: Improving Compositional Reasoning in MLLM Embedding via Reranker Distillation","submittedOnDailyBy":{"_id":"64dc29d9b5d625e0e9a6ecb9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/QxGBsnk1cNsBEPqSx4ae-.jpeg","isPro":false,"fullname":"Tingyu Song","user":"songtingyu","type":"user","name":"songtingyu"},"summary":"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.","upvotes":10,"discussionId":"6a9a278c8f7c3b7557239429","ai_summary":"CORE distills compositional ranking judgments from a cross-attentive reranker into an embedding model via synthesized multi-level candidates and a Rank-KL objective, improving compositional retrieval without degrading standard performance.","ai_keywords":["MLLM-based embedding models","compositional retrieval","cross-attentive reranker","Rank-KL objective","CoSENT","contrastive learning","compositional reasoning benchmarks"],"ai_summary_model":"thinkingmachines/Inkling-Small","organization":{"_id":"661f98de142a51d630dbbcc4","name":"Alibaba-NLP","fullname":"Alibaba-NLP","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/63fc4c00a3c067e62899d32b/dfd_EcIfylvu3sdc2WMqX.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"66af69222f4c59963afc874f","avatarUrl":"/avatars/034ca7688282bdbeddbd4f03e54dead7.svg","isPro":false,"fullname":"Zheyuan Yang","user":"Raywithyou","type":"user"},{"_id":"64dc29d9b5d625e0e9a6ecb9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/QxGBsnk1cNsBEPqSx4ae-.jpeg","isPro":false,"fullname":"Tingyu Song","user":"songtingyu","type":"user"},{"_id":"66e258bdc70c02b46dfed6e3","avatarUrl":"/avatars/ccc2d604616c018f45a268a610472cac.svg","isPro":false,"fullname":"Yuzheng Cai","user":"Ucreate","type":"user"},{"_id":"6434c530a5aed21dd119a393","avatarUrl":"/avatars/a5aa3d8dc8b3b987e6de39280a4c0765.svg","isPro":false,"fullname":"Bro","user":"H34lthy","type":"user"},{"_id":"6a2da6c8ca070ee12c6e396c","avatarUrl":"/avatars/0355287dcabaa67dbc7f0b10b87451f9.svg","isPro":false,"fullname":"Joe Mama","user":"JoeMama123123123","type":"user"},{"_id":"6746b1e2224b22ef67fbff11","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6746b1e2224b22ef67fbff11/q6sN8lQjtLGCgen_41VxU.jpeg","isPro":false,"fullname":"Zhuoning Guo","user":"Zhuoning","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"65dfeee3d16fb170031df293","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65dfeee3d16fb170031df293/2VbNuqcpN3XrWB18NfzRQ.jpeg","isPro":false,"fullname":"gan","user":"guo9","type":"user"},{"_id":"683c642b02c1a474a867964e","avatarUrl":"/avatars/63e44a9cf788ee7b3ad236407700ceca.svg","isPro":false,"fullname":"Jinbiao Wei","user":"mikeweii","type":"user"},{"_id":"684d57f26e04c265777ead3f","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/cuOj-bQqukSZreXgUJlfm.png","isPro":false,"fullname":"Joakim Lee","user":"Reinforcement4All","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"661f98de142a51d630dbbcc4","name":"Alibaba-NLP","fullname":"Alibaba-NLP","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/63fc4c00a3c067e62899d32b/dfd_EcIfylvu3sdc2WMqX.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2609/2609.04083.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2609.04083 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2609.04083 in a Space README.md to link it from this page.