{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:4MHGVD6XTW2WL5MDVON2E56EE7","short_pith_number":"pith:4MHGVD6X","schema_version":"1.0","canonical_sha256":"e30e6a8fd79db565f583ab9ba277c427daa1f8a57e062a0b075b7ac6bd042c17","source":{"kind":"arxiv","id":"2605.29860","version":1},"attestation_state":"computed","paper":{"title":"ESPO: Early-Stopping Proximal Policy Optimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Binhua Li, Jieping Ye, Rui Zhou, Tong Yang, Wenhan Yu, Yingcheng Shi, Yongbin Li, Zeming Li, Zhewen Tan, Zihang Li, Zixiang Liu","submitted_at":"2026-05-28T12:40:22Z","abstract_excerpt":"When a large language model under reinforcement learning commits a wrong reasoning step early in a trajectory, standard algorithms force it to keep generating until the maximum horizon, spending compute on tokens that never receive positive reward and polluting advantage estimates with post-failure noise. We propose ESPO (Early-Stopping Proximal Policy Optimization), which detects trajectory failure on-the-fly and terminates rollouts early. At each generation step, ESPO computes a surrogate regret using only the logits already computed during sampling, and terminates when the smoothed cumulati"},"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":"2605.29860","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-28T12:40:22Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"760400c9737ed52fdd60eaff456f27b525f94f56c826480947450ac3f9aa18f5","abstract_canon_sha256":"9517ae5873ac93c95187f1676cb8d1728316a8c905e568982f01ba53661afb46"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-29T02:05:56.557999Z","signature_b64":"f/TDiExwKTngQidq0w5L8OHWeFobmdm08jLP+YHUysZI2B/D69i6zOvfZs0xhi8bVSE370j39KLy0nbDbWVSDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e30e6a8fd79db565f583ab9ba277c427daa1f8a57e062a0b075b7ac6bd042c17","last_reissued_at":"2026-05-29T02:05:56.557375Z","signature_status":"signed_v1","first_computed_at":"2026-05-29T02:05:56.557375Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ESPO: Early-Stopping Proximal Policy Optimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Binhua Li, Jieping Ye, Rui Zhou, Tong Yang, Wenhan Yu, Yingcheng Shi, Yongbin Li, Zeming Li, Zhewen Tan, Zihang Li, Zixiang Liu","submitted_at":"2026-05-28T12:40:22Z","abstract_excerpt":"When a large language model under reinforcement learning commits a wrong reasoning step early in a trajectory, standard algorithms force it to keep generating until the maximum horizon, spending compute on tokens that never receive positive reward and polluting advantage estimates with post-failure noise. We propose ESPO (Early-Stopping Proximal Policy Optimization), which detects trajectory failure on-the-fly and terminates rollouts early. At each generation step, ESPO computes a surrogate regret using only the logits already computed during sampling, and terminates when the smoothed cumulati"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.29860","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/2605.29860/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":"2605.29860","created_at":"2026-05-29T02:05:56.557481+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.29860v1","created_at":"2026-05-29T02:05:56.557481+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.29860","created_at":"2026-05-29T02:05:56.557481+00:00"},{"alias_kind":"pith_short_12","alias_value":"4MHGVD6XTW2W","created_at":"2026-05-29T02:05:56.557481+00:00"},{"alias_kind":"pith_short_16","alias_value":"4MHGVD6XTW2WL5MD","created_at":"2026-05-29T02:05:56.557481+00:00"},{"alias_kind":"pith_short_8","alias_value":"4MHGVD6X","created_at":"2026-05-29T02:05:56.557481+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/4MHGVD6XTW2WL5MDVON2E56EE7","json":"https://pith.science/pith/4MHGVD6XTW2WL5MDVON2E56EE7.json","graph_json":"https://pith.science/api/pith-number/4MHGVD6XTW2WL5MDVON2E56EE7/graph.json","events_json":"https://pith.science/api/pith-number/4MHGVD6XTW2WL5MDVON2E56EE7/events.json","paper":"https://pith.science/paper/4MHGVD6X"},"agent_actions":{"view_html":"https://pith.science/pith/4MHGVD6XTW2WL5MDVON2E56EE7","download_json":"https://pith.science/pith/4MHGVD6XTW2WL5MDVON2E56EE7.json","view_paper":"https://pith.science/paper/4MHGVD6X","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.29860&json=true","fetch_graph":"https://pith.science/api/pith-number/4MHGVD6XTW2WL5MDVON2E56EE7/graph.json","fetch_events":"https://pith.science/api/pith-number/4MHGVD6XTW2WL5MDVON2E56EE7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4MHGVD6XTW2WL5MDVON2E56EE7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4MHGVD6XTW2WL5MDVON2E56EE7/action/storage_attestation","attest_author":"https://pith.science/pith/4MHGVD6XTW2WL5MDVON2E56EE7/action/author_attestation","sign_citation":"https://pith.science/pith/4MHGVD6XTW2WL5MDVON2E56EE7/action/citation_signature","submit_replication":"https://pith.science/pith/4MHGVD6XTW2WL5MDVON2E56EE7/action/replication_record"}},"created_at":"2026-05-29T02:05:56.557481+00:00","updated_at":"2026-05-29T02:05:56.557481+00:00"}