{"paper":{"title":"GRPO-TTA: Test-Time Visual Tuning for Vision-Language Models via GRPO-Driven Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"GRPO-TTA applies GRPO to test-time visual tuning of vision-language models via group-wise policy optimization on unlabeled class candidates, outperforming prior TTA methods especially under natural distribution shifts.","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Hongyuan Zhang, Yuan Yuan, Yujun Li","submitted_at":"2026-05-05T06:23:20Z","abstract_excerpt":"Group Relative Policy Optimization (GRPO) has recently shown strong performance in post-training large language models and vision-language models. It raises a question of whether the GRPO also significantly promotes the test-time adaptation (TTA) of vision language models. In this paper, we propose Group Relative Policy Optimization for Test-Time Adaptation (GRPO-TTA), which adapts GRPO to the TTA setting by reformulating class-specific prompt prediction as a group-wise policy optimization problem. Specifically, we construct output groups by sampling top-K class candidates from CLIP similarity"},"claims":{"count":3,"items":[{"kind":"strongest_claim","text":"GRPO-TTA consistently outperforms existing test-time adaptation methods, with notably larger performance gains under natural distribution shifts.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That constructing output groups by sampling top-K class candidates from CLIP similarity distributions enables effective probability-driven optimization without ground-truth labels, and that the designed alignment and dispersion rewards guide effective visual encoder tuning at test time.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GRPO-TTA applies GRPO to test-time visual tuning of vision-language models via group-wise policy optimization on unlabeled class candidates, outperforming prior TTA methods especially under natural distribution shifts.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"}],"snapshot_sha256":"52e19a8a2cc17da8513b9278e83bc3f3746a5da9120ae4d53f3ca5ff3998271b"},"source":{"id":"2605.03403","kind":"arxiv","version":2},"verdict":{"id":"3545f4e1-c389-4705-a454-9e0b86db7e03","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-07T17:59:29.852352Z","strongest_claim":"GRPO-TTA consistently outperforms existing test-time adaptation methods, with notably larger performance gains under natural distribution shifts.","one_line_summary":"GRPO-TTA applies GRPO to test-time visual tuning of vision-language models via group-wise policy optimization on unlabeled class candidates, outperforming prior TTA methods especially under natural distribution shifts.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That constructing output groups by sampling top-K class candidates from CLIP similarity distributions enables effective probability-driven optimization without ground-truth labels, and that the designed alignment and dispersion rewards guide effective visual encoder tuning at test time.","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.03403/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T13:42:07.424276Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-20T01:01:22.216932Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T15:26:23.986363Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"3df2f4034b60b226084ee939d6eab9b0511ca69b3d54c9a89cb1470e9b8ddb1d"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}