{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:RLZYC2GBW6I6IBTI34OMTPHSKM","short_pith_number":"pith:RLZYC2GB","schema_version":"1.0","canonical_sha256":"8af38168c1b791e40668df1cc9bcf25319335a9a6caf11cd8da7e6e694d738a3","source":{"kind":"arxiv","id":"2605.20278","version":1},"attestation_state":"computed","paper":{"title":"ClaimDiff-RL: Fine-Grained Caption Reinforcement Learning through Visual Claim Comparison","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Haochen Wang, Hongyang Tang, Jiacheng Chen, Rongxin Guo, Shaoxiang Chen, Tianle Li, Xuyang Shen, Yan Ma, Yu Cheng, Yucong Zhou","submitted_at":"2026-05-19T04:39:28Z","abstract_excerpt":"Long-form image captioning exposes a reward granularity problem in RL: captions are judged as whole sequences, while the important errors occur at the level of individual visual claims. A good dense caption should be both faithful and informative, avoiding hallucination without omitting salient details. Yet pairwise preferences, reference-based metrics, and holistic scalar rewards compress these local errors into a single sequence-level signal, obscuring the tradeoff between factuality and coverage. We introduce ClaimDiff-RL, a framework that uses reference-conditioned atomic claim differences"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.20278","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-19T04:39:28Z","cross_cats_sorted":["cs.AI","cs.CV"],"title_canon_sha256":"d4603f743ef3271e0e5cbc32a0c65712a34f8bc5213ba582bd3469561460cc8e","abstract_canon_sha256":"871ad13974628c3d697f55701260be96622b4c287c2f42614c676a05fd917684"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-21T00:04:23.787715Z","signature_b64":"Odscn8JqrfsGzAcBWVh87dq4RueqXxaoiH57le1m3/kfxHJO8Aur66yjCXF9Mdh6LG4cNyBKny5Hg/GVWnHaCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8af38168c1b791e40668df1cc9bcf25319335a9a6caf11cd8da7e6e694d738a3","last_reissued_at":"2026-05-21T00:04:23.787244Z","signature_status":"signed_v1","first_computed_at":"2026-05-21T00:04:23.787244Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ClaimDiff-RL: Fine-Grained Caption Reinforcement Learning through Visual Claim Comparison","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Haochen Wang, Hongyang Tang, Jiacheng Chen, Rongxin Guo, Shaoxiang Chen, Tianle Li, Xuyang Shen, Yan Ma, Yu Cheng, Yucong Zhou","submitted_at":"2026-05-19T04:39:28Z","abstract_excerpt":"Long-form image captioning exposes a reward granularity problem in RL: captions are judged as whole sequences, while the important errors occur at the level of individual visual claims. A good dense caption should be both faithful and informative, avoiding hallucination without omitting salient details. Yet pairwise preferences, reference-based metrics, and holistic scalar rewards compress these local errors into a single sequence-level signal, obscuring the tradeoff between factuality and coverage. We introduce ClaimDiff-RL, a framework that uses reference-conditioned atomic claim differences"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.20278","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.20278/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2605.20278","created_at":"2026-05-21T00:04:23.787307+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.20278v1","created_at":"2026-05-21T00:04:23.787307+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.20278","created_at":"2026-05-21T00:04:23.787307+00:00"},{"alias_kind":"pith_short_12","alias_value":"RLZYC2GBW6I6","created_at":"2026-05-21T00:04:23.787307+00:00"},{"alias_kind":"pith_short_16","alias_value":"RLZYC2GBW6I6IBTI","created_at":"2026-05-21T00:04:23.787307+00:00"},{"alias_kind":"pith_short_8","alias_value":"RLZYC2GB","created_at":"2026-05-21T00:04:23.787307+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RLZYC2GBW6I6IBTI34OMTPHSKM","json":"https://pith.science/pith/RLZYC2GBW6I6IBTI34OMTPHSKM.json","graph_json":"https://pith.science/api/pith-number/RLZYC2GBW6I6IBTI34OMTPHSKM/graph.json","events_json":"https://pith.science/api/pith-number/RLZYC2GBW6I6IBTI34OMTPHSKM/events.json","paper":"https://pith.science/paper/RLZYC2GB"},"agent_actions":{"view_html":"https://pith.science/pith/RLZYC2GBW6I6IBTI34OMTPHSKM","download_json":"https://pith.science/pith/RLZYC2GBW6I6IBTI34OMTPHSKM.json","view_paper":"https://pith.science/paper/RLZYC2GB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.20278&json=true","fetch_graph":"https://pith.science/api/pith-number/RLZYC2GBW6I6IBTI34OMTPHSKM/graph.json","fetch_events":"https://pith.science/api/pith-number/RLZYC2GBW6I6IBTI34OMTPHSKM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RLZYC2GBW6I6IBTI34OMTPHSKM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RLZYC2GBW6I6IBTI34OMTPHSKM/action/storage_attestation","attest_author":"https://pith.science/pith/RLZYC2GBW6I6IBTI34OMTPHSKM/action/author_attestation","sign_citation":"https://pith.science/pith/RLZYC2GBW6I6IBTI34OMTPHSKM/action/citation_signature","submit_replication":"https://pith.science/pith/RLZYC2GBW6I6IBTI34OMTPHSKM/action/replication_record"}},"created_at":"2026-05-21T00:04:23.787307+00:00","updated_at":"2026-05-21T00:04:23.787307+00:00"}