{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TIWRX7BTCFURL7PW573LTH4P4Z","short_pith_number":"pith:TIWRX7BT","schema_version":"1.0","canonical_sha256":"9a2d1bfc33116915fdf6eff6b99f8fe67f254b191e3a8108d2f3b262b38b8bb7","source":{"kind":"arxiv","id":"2505.15201","version":5},"attestation_state":"computed","paper":{"title":"Pass@K Policy Optimization: Solving Harder Reinforcement Learning Problems","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Christian Walder, Deep Karkhanis","submitted_at":"2025-05-21T07:26:36Z","abstract_excerpt":"Reinforcement Learning (RL) algorithms sample multiple n>1 solution attempts for each problem and reward them independently. This optimizes for pass@1 performance and prioritizes the strength of isolated samples at the expense of the diversity and collective utility of sets of samples. This under-utilizes the sampling capacity, limiting exploration and eventual improvement on harder examples. As a fix, we propose Pass-at-k Policy Optimization (PKPO), a transformation on the final rewards which leads to direct optimization of pass@k performance, thus optimizing for sets of samples that maximize"},"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":"2505.15201","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-21T07:26:36Z","cross_cats_sorted":["cs.AI","cs.CL","stat.ML"],"title_canon_sha256":"90d89e7ce72416ed6a2ad23373530035cb46361592633dbe9c0556e84fac5da3","abstract_canon_sha256":"4ec49d561b89fbf40f89fdfe3c5f1fa2d6c62eca4335bca041152819f59c39ff"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-11T01:09:13.385898Z","signature_b64":"zV07tIsMKHhR1c8UhhTHkD2bK5e4ErQ9Eeven/4jq45fw3sLDaxQX270EnVsuTbDfPRE/l04AJDZxMvH2yqkCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9a2d1bfc33116915fdf6eff6b99f8fe67f254b191e3a8108d2f3b262b38b8bb7","last_reissued_at":"2026-06-11T01:09:13.385182Z","signature_status":"signed_v1","first_computed_at":"2026-06-11T01:09:13.385182Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Pass@K Policy Optimization: Solving Harder Reinforcement Learning Problems","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Christian Walder, Deep Karkhanis","submitted_at":"2025-05-21T07:26:36Z","abstract_excerpt":"Reinforcement Learning (RL) algorithms sample multiple n>1 solution attempts for each problem and reward them independently. This optimizes for pass@1 performance and prioritizes the strength of isolated samples at the expense of the diversity and collective utility of sets of samples. This under-utilizes the sampling capacity, limiting exploration and eventual improvement on harder examples. As a fix, we propose Pass-at-k Policy Optimization (PKPO), a transformation on the final rewards which leads to direct optimization of pass@k performance, thus optimizing for sets of samples that maximize"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.15201","kind":"arxiv","version":5},"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/2505.15201/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":"2505.15201","created_at":"2026-06-11T01:09:13.385269+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.15201v5","created_at":"2026-06-11T01:09:13.385269+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.15201","created_at":"2026-06-11T01:09:13.385269+00:00"},{"alias_kind":"pith_short_12","alias_value":"TIWRX7BTCFUR","created_at":"2026-06-11T01:09:13.385269+00:00"},{"alias_kind":"pith_short_16","alias_value":"TIWRX7BTCFURL7PW","created_at":"2026-06-11T01:09:13.385269+00:00"},{"alias_kind":"pith_short_8","alias_value":"TIWRX7BT","created_at":"2026-06-11T01:09:13.385269+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":11,"internal_anchor_count":11,"sample":[{"citing_arxiv_id":"2605.17333","citing_title":"Leveraging Error Diversity in Group Rollouts for Reinforcement Learning","ref_index":27,"is_internal_anchor":true},{"citing_arxiv_id":"2605.21792","citing_title":"Residual Skill Optimization for Text-to-SQL Ensembles","ref_index":37,"is_internal_anchor":true},{"citing_arxiv_id":"2605.20854","citing_title":"Finite-Time Regret Analysis of Retry-Aware Bandits","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11461","citing_title":"Breaking $\\textit{Winner-Takes-All}$: Cooperative Policy Optimization Improves Diverse LLM Reasoning","ref_index":32,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18864","citing_title":"SAGE: Shaping Anchors for Guided Exploration in RLVR of LLMs","ref_index":31,"is_internal_anchor":true},{"citing_arxiv_id":"2605.17333","citing_title":"Leveraging Error Diversity in Group Rollouts for Reinforcement Learning","ref_index":27,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11461","citing_title":"Breaking $\\textit{Winner-Takes-All}$: Cooperative Policy Optimization Improves Diverse LLM Reasoning","ref_index":32,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10716","citing_title":"What should post-training optimize? A test-time scaling law perspective","ref_index":23,"is_internal_anchor":true},{"citing_arxiv_id":"2605.00380","citing_title":"ResRL: Boosting LLM Reasoning via Negative Sample Projection Residual Reinforcement Learning","ref_index":64,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07039","citing_title":"PACEvolve++: Improving Test-time Learning for Evolutionary Search Agents","ref_index":42,"is_internal_anchor":true},{"citing_arxiv_id":"2605.00380","citing_title":"ResRL: Boosting LLM Reasoning via Negative Sample Projection Residual Reinforcement Learning","ref_index":64,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TIWRX7BTCFURL7PW573LTH4P4Z","json":"https://pith.science/pith/TIWRX7BTCFURL7PW573LTH4P4Z.json","graph_json":"https://pith.science/api/pith-number/TIWRX7BTCFURL7PW573LTH4P4Z/graph.json","events_json":"https://pith.science/api/pith-number/TIWRX7BTCFURL7PW573LTH4P4Z/events.json","paper":"https://pith.science/paper/TIWRX7BT"},"agent_actions":{"view_html":"https://pith.science/pith/TIWRX7BTCFURL7PW573LTH4P4Z","download_json":"https://pith.science/pith/TIWRX7BTCFURL7PW573LTH4P4Z.json","view_paper":"https://pith.science/paper/TIWRX7BT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.15201&json=true","fetch_graph":"https://pith.science/api/pith-number/TIWRX7BTCFURL7PW573LTH4P4Z/graph.json","fetch_events":"https://pith.science/api/pith-number/TIWRX7BTCFURL7PW573LTH4P4Z/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TIWRX7BTCFURL7PW573LTH4P4Z/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TIWRX7BTCFURL7PW573LTH4P4Z/action/storage_attestation","attest_author":"https://pith.science/pith/TIWRX7BTCFURL7PW573LTH4P4Z/action/author_attestation","sign_citation":"https://pith.science/pith/TIWRX7BTCFURL7PW573LTH4P4Z/action/citation_signature","submit_replication":"https://pith.science/pith/TIWRX7BTCFURL7PW573LTH4P4Z/action/replication_record"}},"created_at":"2026-06-11T01:09:13.385269+00:00","updated_at":"2026-06-11T01:09:13.385269+00:00"}