{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:26OFLDTLSEV7LRK74ZLXOKQBGI","short_pith_number":"pith:26OFLDTL","schema_version":"1.0","canonical_sha256":"d79c558e6b912bf5c55fe657772a01321ca17288cd8a07ece8817c6b5e2850a7","source":{"kind":"arxiv","id":"2605.29398","version":1},"attestation_state":"computed","paper":{"title":"GDSD: Reinforcement Learning as Guided Denoiser Self-Distillation for Diffusion Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Che Liu, Ilija Bogunovic, Keyue Jiang, Qifang Zhao, Sangwoong Yoon, Xiaohang Tang, Xiaoxiao Xu","submitted_at":"2026-05-28T05:47:40Z","abstract_excerpt":"Reinforcement learning (RL) can be used to improve the policy (denoiser) of diffusion large language models (dLLMs), while being hindered by the intractability of the policy likelihood. A dominant and efficient family of methods replaces the likelihood in standard RL with its evidence lower bound (ELBO), estimated from randomly masked sequences. Despite being well aligned with pre-training, these approaches introduce bias through training--inference mismatch by using the ELBO as a likelihood surrogate, which can degrade performance. In this work, we propose Guided Denoiser Self-Distillation (G"},"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":"2605.29398","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-28T05:47:40Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"dde5a7c57de24e903572613750bf63f443ac80ee37e380f7b6474f992581ad8e","abstract_canon_sha256":"ffbf0ffeaefd483e7f45bb3473c1f7e1473c27f7419bfc15da877e91324039de"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-29T01:05:37.022187Z","signature_b64":"L9kbkS6nZYpXHVLu5f6trxmWHYxt06PKvGQX5O0G9az8N2BaXwRWAifjtutv79CtbK/NPyZ6mCQrNi6RgNiOAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d79c558e6b912bf5c55fe657772a01321ca17288cd8a07ece8817c6b5e2850a7","last_reissued_at":"2026-05-29T01:05:37.021404Z","signature_status":"signed_v1","first_computed_at":"2026-05-29T01:05:37.021404Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"GDSD: Reinforcement Learning as Guided Denoiser Self-Distillation for Diffusion Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Che Liu, Ilija Bogunovic, Keyue Jiang, Qifang Zhao, Sangwoong Yoon, Xiaohang Tang, Xiaoxiao Xu","submitted_at":"2026-05-28T05:47:40Z","abstract_excerpt":"Reinforcement learning (RL) can be used to improve the policy (denoiser) of diffusion large language models (dLLMs), while being hindered by the intractability of the policy likelihood. A dominant and efficient family of methods replaces the likelihood in standard RL with its evidence lower bound (ELBO), estimated from randomly masked sequences. Despite being well aligned with pre-training, these approaches introduce bias through training--inference mismatch by using the ELBO as a likelihood surrogate, which can degrade performance. In this work, we propose Guided Denoiser Self-Distillation (G"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.29398","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/2605.29398/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":"2605.29398","created_at":"2026-05-29T01:05:37.021525+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.29398v1","created_at":"2026-05-29T01:05:37.021525+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.29398","created_at":"2026-05-29T01:05:37.021525+00:00"},{"alias_kind":"pith_short_12","alias_value":"26OFLDTLSEV7","created_at":"2026-05-29T01:05:37.021525+00:00"},{"alias_kind":"pith_short_16","alias_value":"26OFLDTLSEV7LRK7","created_at":"2026-05-29T01:05:37.021525+00:00"},{"alias_kind":"pith_short_8","alias_value":"26OFLDTL","created_at":"2026-05-29T01:05:37.021525+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/26OFLDTLSEV7LRK74ZLXOKQBGI","json":"https://pith.science/pith/26OFLDTLSEV7LRK74ZLXOKQBGI.json","graph_json":"https://pith.science/api/pith-number/26OFLDTLSEV7LRK74ZLXOKQBGI/graph.json","events_json":"https://pith.science/api/pith-number/26OFLDTLSEV7LRK74ZLXOKQBGI/events.json","paper":"https://pith.science/paper/26OFLDTL"},"agent_actions":{"view_html":"https://pith.science/pith/26OFLDTLSEV7LRK74ZLXOKQBGI","download_json":"https://pith.science/pith/26OFLDTLSEV7LRK74ZLXOKQBGI.json","view_paper":"https://pith.science/paper/26OFLDTL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.29398&json=true","fetch_graph":"https://pith.science/api/pith-number/26OFLDTLSEV7LRK74ZLXOKQBGI/graph.json","fetch_events":"https://pith.science/api/pith-number/26OFLDTLSEV7LRK74ZLXOKQBGI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/26OFLDTLSEV7LRK74ZLXOKQBGI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/26OFLDTLSEV7LRK74ZLXOKQBGI/action/storage_attestation","attest_author":"https://pith.science/pith/26OFLDTLSEV7LRK74ZLXOKQBGI/action/author_attestation","sign_citation":"https://pith.science/pith/26OFLDTLSEV7LRK74ZLXOKQBGI/action/citation_signature","submit_replication":"https://pith.science/pith/26OFLDTLSEV7LRK74ZLXOKQBGI/action/replication_record"}},"created_at":"2026-05-29T01:05:37.021525+00:00","updated_at":"2026-05-29T01:05:37.021525+00:00"}