{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:LCB3UHAAN6GMCNM5457GPWNF6W","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"c32ff01e1d574c8407f7d747be89e3134028b09fc84bdb3106f98de9fc65a742","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-04-10T17:41:56Z","title_canon_sha256":"44ab9014d6ac1819ca9747e6da46d768642c25e8940acc1356b7d8608ad8420c"},"schema_version":"1.0","source":{"id":"2504.08837","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2504.08837","created_at":"2026-05-17T23:38:53Z"},{"alias_kind":"arxiv_version","alias_value":"2504.08837v3","created_at":"2026-05-17T23:38:53Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.08837","created_at":"2026-05-17T23:38:53Z"},{"alias_kind":"pith_short_12","alias_value":"LCB3UHAAN6GM","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"LCB3UHAAN6GMCNM5","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"LCB3UHAA","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:12d80398d7d30da12b5805987cae7af41b892d4b8386af16a5b2a87d930c0663","target":"graph","created_at":"2026-05-17T23:38:53Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"By combining Selective Sample Replay and Forced Rethinking in RL training, VL-Rethinker advances state-of-the-art scores on MathVista to 80.4% and MathVerse to 63.5%, achieving open-source SoTA on MathVision, MMMU-Pro, EMMA, and MEGA-Bench."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the reported gains stem primarily from increased self-reflection and slow-thinking rather than from other side effects of the RL setup or from benchmark-specific optimizations."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"VL-Rethinker reaches 80.4% on MathVista and 63.5% on MathVerse by adapting GRPO with Selective Sample Replay and Forced Rethinking to promote self-reflection in vision-language models without distillation."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Reinforcement learning with selective replay and forced rethinking steps lets vision-language models reflect on their answers and reach new highs on multimodal math benchmarks."}],"snapshot_sha256":"b0c4414d16041412c92919d0cf6aa837c74158d31ffffaf3905eb35a98691f03"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"8ceeed42e106089115381c36b37aff4df123d475d9489a7c85c2ad5702295eae"},"paper":{"abstract_excerpt":"Recently, slow-thinking systems like GPT-o1 and DeepSeek-R1 have demonstrated great potential in solving challenging problems through explicit reflection. They significantly outperform the best fast-thinking models, such as GPT-4o, on various math and science benchmarks. However, their multimodal reasoning capabilities remain on par with fast-thinking models. For instance, GPT-o1's performance on benchmarks like MathVista, MathVerse, and MathVision is similar to fast-thinking models. In this paper, we aim to enhance the slow-thinking capabilities of vision-language models using reinforcement l","authors_text":"Chao Qu, Fangzhen Lin, Haozhe Wang, Wei Chu, Wenhu Chen, Zuming Huang","cross_cats":["cs.AI"],"headline":"Reinforcement learning with selective replay and forced rethinking steps lets vision-language models reflect on their answers and reach new highs on multimodal math benchmarks.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-04-10T17:41:56Z","title":"VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.08837","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-15T06:08:53.147112Z","id":"fb142a08-24f6-4a45-93ca-6e6124f8f84e","model_set":{"reader":"grok-4.3"},"one_line_summary":"VL-Rethinker reaches 80.4% on MathVista and 63.5% on MathVerse by adapting GRPO with Selective Sample Replay and Forced Rethinking to promote self-reflection in vision-language models without distillation.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Reinforcement learning with selective replay and forced rethinking steps lets vision-language models reflect on their answers and reach new highs on multimodal math benchmarks.","strongest_claim":"By combining Selective Sample Replay and Forced Rethinking in RL training, VL-Rethinker advances state-of-the-art scores on MathVista to 80.4% and MathVerse to 63.5%, achieving open-source SoTA on MathVision, MMMU-Pro, EMMA, and MEGA-Bench.","weakest_assumption":"That the reported gains stem primarily from increased self-reflection and slow-thinking rather than from other side effects of the RL setup or from benchmark-specific optimizations."}},"verdict_id":"fb142a08-24f6-4a45-93ca-6e6124f8f84e"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:d6e1a7ea5ea0f9723161ac381e965f4e5edc60d01add0b8083a78e46e37a03f7","target":"record","created_at":"2026-05-17T23:38:53Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"c32ff01e1d574c8407f7d747be89e3134028b09fc84bdb3106f98de9fc65a742","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-04-10T17:41:56Z","title_canon_sha256":"44ab9014d6ac1819ca9747e6da46d768642c25e8940acc1356b7d8608ad8420c"},"schema_version":"1.0","source":{"id":"2504.08837","kind":"arxiv","version":3}},"canonical_sha256":"5883ba1c006f8cc1359de77e67d9a5f5b64427899b130d9f6a57ce29a199434e","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"5883ba1c006f8cc1359de77e67d9a5f5b64427899b130d9f6a57ce29a199434e","first_computed_at":"2026-05-17T23:38:53.290891Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:53.290891Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"0Ff4uNfcoPDKTyfR1lBguygYU6b7kNM58yY0mEe2JXdKmOJJklcUWxzdRnibryfzJEeeZeftfkMrsKwyVQNiAw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:53.291529Z","signed_message":"canonical_sha256_bytes"},"source_id":"2504.08837","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:d6e1a7ea5ea0f9723161ac381e965f4e5edc60d01add0b8083a78e46e37a03f7","sha256:12d80398d7d30da12b5805987cae7af41b892d4b8386af16a5b2a87d930c0663"],"state_sha256":"eec785480a54440773d7d655c8dcbc4e00ebd1fbc3229d16d3d95114d391109f"}