An effective way for resolving conflict in multi-source visual reasoning</p>\n","updatedAt":"2026-05-27T02:13:56.844Z","author":{"_id":"667b7388580a979b703c3c2b","avatarUrl":"/avatars/15d42bf75f88f1e2ac112c98cb53bad6.svg","fullname":"Zeng","name":"AuroraZengfh","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8792248964309692},"editors":["AuroraZengfh"],"editorAvatarUrls":["/avatars/15d42bf75f88f1e2ac112c98cb53bad6.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.25437","authors":[{"_id":"6a165281e9aa3c8e322db32d","user":{"_id":"667b7388580a979b703c3c2b","avatarUrl":"/avatars/15d42bf75f88f1e2ac112c98cb53bad6.svg","isPro":false,"fullname":"Zeng","user":"AuroraZengfh","type":"user","name":"AuroraZengfh"},"name":"Fanhu Zeng","status":"claimed_verified","statusLastChangedAt":"2026-05-27T07:42:07.253Z","hidden":false},{"_id":"6a165281e9aa3c8e322db32e","name":"Zhicong Luo","hidden":false},{"_id":"6a165281e9aa3c8e322db32f","name":"Zefan Wang","hidden":false},{"_id":"6a165281e9aa3c8e322db330","name":"You Li","hidden":false},{"_id":"6a165281e9aa3c8e322db331","name":"Chi Chen","hidden":false},{"_id":"6a165281e9aa3c8e322db332","name":"Maosong Sun","hidden":false}],"publishedAt":"2026-05-25T00:00:00.000Z","submittedOnDailyAt":"2026-05-27T00:00:00.000Z","title":"Does Seeing More Mean Knowing More? Mono-Anchored Advantage Normalization for Multi-Source Visual Reasoning","submittedOnDailyBy":{"_id":"667b7388580a979b703c3c2b","avatarUrl":"/avatars/15d42bf75f88f1e2ac112c98cb53bad6.svg","isPro":false,"fullname":"Zeng","user":"AuroraZengfh","type":"user","name":"AuroraZengfh"},"summary":"Visual reasoning through reinforcement learning with verifiable rewards (RLVR) has achieved remarkable progress. However, when dealing with multi-source inputs, existing approaches tend to treat them as a mere accumulation of information, lacking explicit mechanisms to distinguish whether integrating additional sources yields information gain or introduces interference. Therefore, they struggle to effectively model dynamic interaction when integrating multiple sources, particularly when they differ significantly in physical properties and semantics, e.g., infrared and depth, leading to inferior performance to mono-source reasoning when a certain source holds the dominant signal. To address this issue, we propose MARS, a novel mono-anchored multi-source reasoning framework that models each visual modality as an independent information source. Specifically, by treating mono-source rewards as dynamic anchors, our method explicitly incorporates the information gain introduced by multi-source fusion into advantage normalization and adaptively emphasizes mutual promotion between sources while suppressing potential noise or conflicts during RLVR. From theoretical analysis, our method effectively quantifies information gain introduced by multi-source integration in gradient estimation, enabling consistent modality regulation. Empirical results also show impressive 3.2% and 4.9% performance gains on GRPO and DAPO across diverse datasets, confirming effectiveness of our method.","upvotes":9,"discussionId":"6a165282e9aa3c8e322db333","githubRepo":"https://github.com/AI9Stars/MARS","githubRepoAddedBy":"user","ai_summary":"A novel mono-anchored multi-source reasoning framework that uses dynamic anchors to quantify information gain and regulate modality interactions during reinforcement learning with verifiable rewards.","ai_keywords":["reinforcement learning","visual reasoning","verifiable rewards","multi-source inputs","information gain","advantage normalization","gradient estimation","modality regulation"],"githubStars":2,"organization":{"_id":"628735cbc83a2d6ab8d14a66","name":"Tsinghua","fullname":"Tsinghua University","avatar":"https://www.gravatar.com/avatar/6c5c1441e3283e7543342e59277ea219?d=retro&size=100"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"667b7388580a979b703c3c2b","avatarUrl":"/avatars/15d42bf75f88f1e2ac112c98cb53bad6.svg","isPro":false,"fullname":"Zeng","user":"AuroraZengfh","type":"user"},{"_id":"642086ed290342c5df85662d","avatarUrl":"/avatars/915a4d7b89455ae97b8544c79286ddf8.svg","isPro":false,"fullname":"Chi Chen","user":"carboncoo","type":"user"},{"_id":"654f3e104c8874c64d43aafa","avatarUrl":"/avatars/00de263f98a81c52cdb321fb11b16c06.svg","isPro":false,"fullname":"You Li","user":"Michael4933","type":"user"},{"_id":"65903c4aa78a277803bde77b","avatarUrl":"/avatars/061388320ccf1a66e1e99519dd426a60.svg","isPro":false,"fullname":"Ziyang Ding","user":"Oscar-dzy","type":"user"},{"_id":"6617600c843718c34d151de9","avatarUrl":"/avatars/22fa64949780da15e91d77387bdf1e41.svg","isPro":false,"fullname":"dai","user":"daisq","type":"user"},{"_id":"6582c3cc1c4454dde6e5eb3e","avatarUrl":"/avatars/e4de12207e19a79ec424048da34ebd39.svg","isPro":false,"fullname":"Wang","user":"wph6","type":"user"},{"_id":"68cc3513b13e69da2105147c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/v4F7z_mNQc6CHGIw3sdma.png","isPro":false,"fullname":"liok","user":"michaelpopo","type":"user"},{"_id":"6785c5de58328c47557dcd94","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/QgWM1f07Srpea8ySGDVaK.png","isPro":false,"fullname":"one","user":"bigcg","type":"user"},{"_id":"688350951625a653c117e24d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/lg7LX14oS-JgxlcSOZmm-.jpeg","isPro":false,"fullname":"蔡静楠","user":"caijingnan","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"628735cbc83a2d6ab8d14a66","name":"Tsinghua","fullname":"Tsinghua University","avatar":"https://www.gravatar.com/avatar/6c5c1441e3283e7543342e59277ea219?d=retro&size=100"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.25437.md"}">
Does Seeing More Mean Knowing More? Mono-Anchored Advantage Normalization for Multi-Source Visual Reasoning
Published on May 25
· Submitted by Zeng on May 27 Abstract
A novel mono-anchored multi-source reasoning framework that uses dynamic anchors to quantify information gain and regulate modality interactions during reinforcement learning with verifiable rewards.
AI-generated summary
Visual reasoning through reinforcement learning with verifiable rewards (RLVR) has achieved remarkable progress. However, when dealing with multi-source inputs, existing approaches tend to treat them as a mere accumulation of information, lacking explicit mechanisms to distinguish whether integrating additional sources yields information gain or introduces interference. Therefore, they struggle to effectively model dynamic interaction when integrating multiple sources, particularly when they differ significantly in physical properties and semantics, e.g., infrared and depth, leading to inferior performance to mono-source reasoning when a certain source holds the dominant signal. To address this issue, we propose MARS, a novel mono-anchored multi-source reasoning framework that models each visual modality as an independent information source. Specifically, by treating mono-source rewards as dynamic anchors, our method explicitly incorporates the information gain introduced by multi-source fusion into advantage normalization and adaptively emphasizes mutual promotion between sources while suppressing potential noise or conflicts during RLVR. From theoretical analysis, our method effectively quantifies information gain introduced by multi-source integration in gradient estimation, enabling consistent modality regulation. Empirical results also show impressive 3.2% and 4.9% performance gains on GRPO and DAPO across diverse datasets, confirming effectiveness of our method.
Community
An effective way for resolving conflict in multi-source visual reasoning
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2605.25437 in a model README.md to link it from this page.
Cite arxiv.org/abs/2605.25437 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2605.25437 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.