{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:VYRRYU5V2OEKDYPWCQYRMNVAL3","short_pith_number":"pith:VYRRYU5V","schema_version":"1.0","canonical_sha256":"ae231c53b5d388a1e1f614311636a05ee61d2ad704ddaa61003708e5dbacef0e","source":{"kind":"arxiv","id":"2606.03620","version":1},"attestation_state":"computed","paper":{"title":"Physics-Guided Policy Optimization with Self-Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chaoqun Jia, Devin Chen, Haoran Liu, Kai Wei, Ke Wang, Yuning Wu","submitted_at":"2026-06-02T13:20:39Z","abstract_excerpt":"Self-distilled policy optimization (SDPO) has become a popular paradigm for LLM post-training, where a model learns from its own predictions conditioned on privileged information. SDPO, however, is sensitive to how much each update step should be trusted: corrections from a self-teacher can be highly informative on some batches and misleading on others, and applying them uniformly with a fixed step size can destabilize training. Drawing inspiration from viscous-fluid dynamics and formalizing the analogy at the SDE level, we propose Physics-Guided Policy Optimization (PGPO), which introduces an"},"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":"2606.03620","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-02T13:20:39Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"7517ea9569d5c8ba3e685a6e043df15829e3b434f436d9b23ce412d245edcbb5","abstract_canon_sha256":"7fe97636fab17b4355f8387a78ffe7816a48427627266cce0d06efe33a1316c0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-03T01:06:02.527647Z","signature_b64":"2TdNFuDuBiK9AYjL4ztB8BkBeOzaUed2C1pXaLmjBncTHFDG8taavCGzTHo6h2eRDeV05gXjUkur3dv7y4oWDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ae231c53b5d388a1e1f614311636a05ee61d2ad704ddaa61003708e5dbacef0e","last_reissued_at":"2026-06-03T01:06:02.527290Z","signature_status":"signed_v1","first_computed_at":"2026-06-03T01:06:02.527290Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Physics-Guided Policy Optimization with Self-Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chaoqun Jia, Devin Chen, Haoran Liu, Kai Wei, Ke Wang, Yuning Wu","submitted_at":"2026-06-02T13:20:39Z","abstract_excerpt":"Self-distilled policy optimization (SDPO) has become a popular paradigm for LLM post-training, where a model learns from its own predictions conditioned on privileged information. SDPO, however, is sensitive to how much each update step should be trusted: corrections from a self-teacher can be highly informative on some batches and misleading on others, and applying them uniformly with a fixed step size can destabilize training. Drawing inspiration from viscous-fluid dynamics and formalizing the analogy at the SDE level, we propose Physics-Guided Policy Optimization (PGPO), which introduces an"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.03620","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/2606.03620/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":"2606.03620","created_at":"2026-06-03T01:06:02.527358+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.03620v1","created_at":"2026-06-03T01:06:02.527358+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.03620","created_at":"2026-06-03T01:06:02.527358+00:00"},{"alias_kind":"pith_short_12","alias_value":"VYRRYU5V2OEK","created_at":"2026-06-03T01:06:02.527358+00:00"},{"alias_kind":"pith_short_16","alias_value":"VYRRYU5V2OEKDYPW","created_at":"2026-06-03T01:06:02.527358+00:00"},{"alias_kind":"pith_short_8","alias_value":"VYRRYU5V","created_at":"2026-06-03T01:06:02.527358+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2606.30345","citing_title":"DRIFT: Difficulty Routing Self-DIstillation with Rhythm-Gated Exploration and Success BuFfer Training","ref_index":19,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VYRRYU5V2OEKDYPWCQYRMNVAL3","json":"https://pith.science/pith/VYRRYU5V2OEKDYPWCQYRMNVAL3.json","graph_json":"https://pith.science/api/pith-number/VYRRYU5V2OEKDYPWCQYRMNVAL3/graph.json","events_json":"https://pith.science/api/pith-number/VYRRYU5V2OEKDYPWCQYRMNVAL3/events.json","paper":"https://pith.science/paper/VYRRYU5V"},"agent_actions":{"view_html":"https://pith.science/pith/VYRRYU5V2OEKDYPWCQYRMNVAL3","download_json":"https://pith.science/pith/VYRRYU5V2OEKDYPWCQYRMNVAL3.json","view_paper":"https://pith.science/paper/VYRRYU5V","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.03620&json=true","fetch_graph":"https://pith.science/api/pith-number/VYRRYU5V2OEKDYPWCQYRMNVAL3/graph.json","fetch_events":"https://pith.science/api/pith-number/VYRRYU5V2OEKDYPWCQYRMNVAL3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VYRRYU5V2OEKDYPWCQYRMNVAL3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VYRRYU5V2OEKDYPWCQYRMNVAL3/action/storage_attestation","attest_author":"https://pith.science/pith/VYRRYU5V2OEKDYPWCQYRMNVAL3/action/author_attestation","sign_citation":"https://pith.science/pith/VYRRYU5V2OEKDYPWCQYRMNVAL3/action/citation_signature","submit_replication":"https://pith.science/pith/VYRRYU5V2OEKDYPWCQYRMNVAL3/action/replication_record"}},"created_at":"2026-06-03T01:06:02.527358+00:00","updated_at":"2026-06-03T01:06:02.527358+00:00"}