{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:IJHYJWCC7YP6LJPF5AS6DOJA2O","short_pith_number":"pith:IJHYJWCC","schema_version":"1.0","canonical_sha256":"424f84d842fe1fe5a5e5e825e1b920d39bf13dae59ca10da968f53356848d33e","source":{"kind":"arxiv","id":"2605.19461","version":1},"attestation_state":"computed","paper":{"title":"Beyond Mode Collapse: Distribution Matching for Diverse Reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Dahua Lin, Kai Chen, Linyang Li, Peiji Li, Qingwen Liu, Qipeng Guo, Shengyuan Ding, Tianyi Lyu, Xiaozhe Li, Xinyu Fang, Yang Li, Yichuan Ma, Yongkang Chen","submitted_at":"2026-05-19T07:13:00Z","abstract_excerpt":"On-policy reinforcement learning methods like GRPO suffer from mode collapse: they exhibit reduced solution diversity, concentrating probability mass on a single solution once discovered and ceasing exploration of alternative strategies. We show this stems from reverse KL minimization's mode-seeking behavior, which reinforces the first high-reward trajectory found rather than maintaining a distribution over multiple diverse solutions. We propose DMPO (Distribution-Matching Policy Optimization), which prevents mode collapse through principled approximation of forward KL minimization. DMPO const"},"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":"2605.19461","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-19T07:13:00Z","cross_cats_sorted":[],"title_canon_sha256":"89fd0952555b7ba2d5f22df70deffaba80a12f66653fb1ca728e59c387683eb5","abstract_canon_sha256":"f1c2984de532e1821df695509ac3a9632fdf144fa7df790d1bf87639459cda76"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T01:05:47.032402Z","signature_b64":"ge8VR5XRcOuUUC9Ru/We0f8Dtr5its2HmyNkAJBlHahIHm5M+J/zVbwWp+v3SOSfq5hnWSzj2ZdIO9VfaBsMCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"424f84d842fe1fe5a5e5e825e1b920d39bf13dae59ca10da968f53356848d33e","last_reissued_at":"2026-05-20T01:05:47.031649Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T01:05:47.031649Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Beyond Mode Collapse: Distribution Matching for Diverse Reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Dahua Lin, Kai Chen, Linyang Li, Peiji Li, Qingwen Liu, Qipeng Guo, Shengyuan Ding, Tianyi Lyu, Xiaozhe Li, Xinyu Fang, Yang Li, Yichuan Ma, Yongkang Chen","submitted_at":"2026-05-19T07:13:00Z","abstract_excerpt":"On-policy reinforcement learning methods like GRPO suffer from mode collapse: they exhibit reduced solution diversity, concentrating probability mass on a single solution once discovered and ceasing exploration of alternative strategies. We show this stems from reverse KL minimization's mode-seeking behavior, which reinforces the first high-reward trajectory found rather than maintaining a distribution over multiple diverse solutions. We propose DMPO (Distribution-Matching Policy Optimization), which prevents mode collapse through principled approximation of forward KL minimization. DMPO const"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.19461","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/2605.19461/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":"2605.19461","created_at":"2026-05-20T01:05:47.031775+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.19461v1","created_at":"2026-05-20T01:05:47.031775+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.19461","created_at":"2026-05-20T01:05:47.031775+00:00"},{"alias_kind":"pith_short_12","alias_value":"IJHYJWCC7YP6","created_at":"2026-05-20T01:05:47.031775+00:00"},{"alias_kind":"pith_short_16","alias_value":"IJHYJWCC7YP6LJPF","created_at":"2026-05-20T01:05:47.031775+00:00"},{"alias_kind":"pith_short_8","alias_value":"IJHYJWCC","created_at":"2026-05-20T01:05:47.031775+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/IJHYJWCC7YP6LJPF5AS6DOJA2O","json":"https://pith.science/pith/IJHYJWCC7YP6LJPF5AS6DOJA2O.json","graph_json":"https://pith.science/api/pith-number/IJHYJWCC7YP6LJPF5AS6DOJA2O/graph.json","events_json":"https://pith.science/api/pith-number/IJHYJWCC7YP6LJPF5AS6DOJA2O/events.json","paper":"https://pith.science/paper/IJHYJWCC"},"agent_actions":{"view_html":"https://pith.science/pith/IJHYJWCC7YP6LJPF5AS6DOJA2O","download_json":"https://pith.science/pith/IJHYJWCC7YP6LJPF5AS6DOJA2O.json","view_paper":"https://pith.science/paper/IJHYJWCC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.19461&json=true","fetch_graph":"https://pith.science/api/pith-number/IJHYJWCC7YP6LJPF5AS6DOJA2O/graph.json","fetch_events":"https://pith.science/api/pith-number/IJHYJWCC7YP6LJPF5AS6DOJA2O/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IJHYJWCC7YP6LJPF5AS6DOJA2O/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IJHYJWCC7YP6LJPF5AS6DOJA2O/action/storage_attestation","attest_author":"https://pith.science/pith/IJHYJWCC7YP6LJPF5AS6DOJA2O/action/author_attestation","sign_citation":"https://pith.science/pith/IJHYJWCC7YP6LJPF5AS6DOJA2O/action/citation_signature","submit_replication":"https://pith.science/pith/IJHYJWCC7YP6LJPF5AS6DOJA2O/action/replication_record"}},"created_at":"2026-05-20T01:05:47.031775+00:00","updated_at":"2026-05-20T01:05:47.031775+00:00"}