{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:GA3KHFDVH3EOIQAVX2MHKM627T","short_pith_number":"pith:GA3KHFDV","schema_version":"1.0","canonical_sha256":"3036a394753ec8e44015be987533dafcde8eafa8253a83fab5258f0d62d3307c","source":{"kind":"arxiv","id":"2605.26013","version":1},"attestation_state":"computed","paper":{"title":"AdvantageFlow: Advantage-Weighted Least Squares for RL in Flow Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Anup Rao, Branislav Kveton, Krishna Kumar Singh, Subhojyoti Mukherjee, Viet Dac Lai","submitted_at":"2026-05-25T16:32:14Z","abstract_excerpt":"We introduce AdvantageFlow, a forward-process reinforcement learning algorithm for rectified flow models. Unlike Flow-GRPO, which optimizes the reverse process, we optimize an advantage-weighted forward-process prediction loss. This optimization problem is unstable when advantages are negative and the loss becomes non-convex. We stabilize it by rollout policy regularization, which reduces variance and arises from fitting a local reward-improving target distribution. We evaluate AdvantageFlow on image generation tasks with Stable Diffusion 3.5 Medium. It outperforms both Flow-GRPO and a state-o"},"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":"2605.26013","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-25T16:32:14Z","cross_cats_sorted":["cs.AI","cs.CV"],"title_canon_sha256":"79c00d175876ebf62899fb383bd168913498fd290e7dc9daea278a01c71ac481","abstract_canon_sha256":"24ef5696f1103e19d8de1fbd5aa567d49dc4a04bf4201416f90c07223141e395"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T02:05:23.267626Z","signature_b64":"2yv4u7xcTepPZAG3aCYIRTTj1WWT2jCcS945/VRWNzVva/de3+nOaMVBjmbzaxXOEDChKjW2DB4g3ZzBgx2ECg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3036a394753ec8e44015be987533dafcde8eafa8253a83fab5258f0d62d3307c","last_reissued_at":"2026-05-26T02:05:23.266900Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T02:05:23.266900Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"AdvantageFlow: Advantage-Weighted Least Squares for RL in Flow Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Anup Rao, Branislav Kveton, Krishna Kumar Singh, Subhojyoti Mukherjee, Viet Dac Lai","submitted_at":"2026-05-25T16:32:14Z","abstract_excerpt":"We introduce AdvantageFlow, a forward-process reinforcement learning algorithm for rectified flow models. Unlike Flow-GRPO, which optimizes the reverse process, we optimize an advantage-weighted forward-process prediction loss. This optimization problem is unstable when advantages are negative and the loss becomes non-convex. We stabilize it by rollout policy regularization, which reduces variance and arises from fitting a local reward-improving target distribution. We evaluate AdvantageFlow on image generation tasks with Stable Diffusion 3.5 Medium. It outperforms both Flow-GRPO and a state-o"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.26013","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/2605.26013/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":"2605.26013","created_at":"2026-05-26T02:05:23.267023+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.26013v1","created_at":"2026-05-26T02:05:23.267023+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.26013","created_at":"2026-05-26T02:05:23.267023+00:00"},{"alias_kind":"pith_short_12","alias_value":"GA3KHFDVH3EO","created_at":"2026-05-26T02:05:23.267023+00:00"},{"alias_kind":"pith_short_16","alias_value":"GA3KHFDVH3EOIQAV","created_at":"2026-05-26T02:05:23.267023+00:00"},{"alias_kind":"pith_short_8","alias_value":"GA3KHFDV","created_at":"2026-05-26T02:05:23.267023+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/GA3KHFDVH3EOIQAVX2MHKM627T","json":"https://pith.science/pith/GA3KHFDVH3EOIQAVX2MHKM627T.json","graph_json":"https://pith.science/api/pith-number/GA3KHFDVH3EOIQAVX2MHKM627T/graph.json","events_json":"https://pith.science/api/pith-number/GA3KHFDVH3EOIQAVX2MHKM627T/events.json","paper":"https://pith.science/paper/GA3KHFDV"},"agent_actions":{"view_html":"https://pith.science/pith/GA3KHFDVH3EOIQAVX2MHKM627T","download_json":"https://pith.science/pith/GA3KHFDVH3EOIQAVX2MHKM627T.json","view_paper":"https://pith.science/paper/GA3KHFDV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.26013&json=true","fetch_graph":"https://pith.science/api/pith-number/GA3KHFDVH3EOIQAVX2MHKM627T/graph.json","fetch_events":"https://pith.science/api/pith-number/GA3KHFDVH3EOIQAVX2MHKM627T/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GA3KHFDVH3EOIQAVX2MHKM627T/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GA3KHFDVH3EOIQAVX2MHKM627T/action/storage_attestation","attest_author":"https://pith.science/pith/GA3KHFDVH3EOIQAVX2MHKM627T/action/author_attestation","sign_citation":"https://pith.science/pith/GA3KHFDVH3EOIQAVX2MHKM627T/action/citation_signature","submit_replication":"https://pith.science/pith/GA3KHFDVH3EOIQAVX2MHKM627T/action/replication_record"}},"created_at":"2026-05-26T02:05:23.267023+00:00","updated_at":"2026-05-26T02:05:23.267023+00:00"}