{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:LWK7VBFA7CBQDKVKVILWGL6A57","short_pith_number":"pith:LWK7VBFA","schema_version":"1.0","canonical_sha256":"5d95fa84a0f88301aaaaaa17632fc0eff9c44a5eb7e213828739691fad4c1246","source":{"kind":"arxiv","id":"2604.24357","version":2},"attestation_state":"computed","paper":{"title":"DPRM: A Plug-in Doob h transform-induced Token-Ordering Module for Diffusion Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"DPRM introduces a plug-in module that shifts token ordering in diffusion language models from confidence rules to Doob h-transform process reward guidance.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Andi Han, Atsushi Nitanda, Dake Bu, Hau-San Wong, Qingfu Zhang, Taiji Suzuki, Wei Huang","submitted_at":"2026-04-27T11:50:26Z","abstract_excerpt":"Diffusion language models generate without a fixed left-to-right order, leaving token ordering as a central algorithmic choice. Existing systems mainly use random masking or confidence-driven ordering, which respectively suffer from train--test mismatch and myopic exploration. We introduce DPRM (Doob -transform Process Reward Model), a plug-in token-ordering module that keeps the host architecture, denoising objective and supervision unchanged, and modifies only the ordering policy. DPRM starts from confidence-driven ordering and gradually shifts to process-reward-guided ordering through onlin"},"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":"2604.24357","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-27T11:50:26Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"49042bc26768a5363f48d7758d734ea584e7de7c8e22c1743c4ce050639a39ed","abstract_canon_sha256":"071a7288cc5839152eda0dfaf391407930a80f73c3b18801f55879ce8b95ff49"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-19T16:09:58.559831Z","signature_b64":"QjsEj+f0hvqTGdygxkXaleIheLHbyFFjJAN2V0b18NsLJ+ju/I/mq42iOucbYEvSVnluKF6tqOcyWgwYgWI7Bg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5d95fa84a0f88301aaaaaa17632fc0eff9c44a5eb7e213828739691fad4c1246","last_reissued_at":"2026-06-19T16:09:58.559402Z","signature_status":"signed_v1","first_computed_at":"2026-06-19T16:09:58.559402Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DPRM: A Plug-in Doob h transform-induced Token-Ordering Module for Diffusion Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"DPRM introduces a plug-in module that shifts token ordering in diffusion language models from confidence rules to Doob h-transform process reward guidance.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Andi Han, Atsushi Nitanda, Dake Bu, Hau-San Wong, Qingfu Zhang, Taiji Suzuki, Wei Huang","submitted_at":"2026-04-27T11:50:26Z","abstract_excerpt":"Diffusion language models generate without a fixed left-to-right order, leaving token ordering as a central algorithmic choice. Existing systems mainly use random masking or confidence-driven ordering, which respectively suffer from train--test mismatch and myopic exploration. We introduce DPRM (Doob -transform Process Reward Model), a plug-in token-ordering module that keeps the host architecture, denoising objective and supervision unchanged, and modifies only the ordering policy. DPRM starts from confidence-driven ordering and gradually shifts to process-reward-guided ordering through onlin"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"DPRM improves over confidence-based baselines in pretraining, post-training, test-time scaling, and single-cell masked diffusion, with particularly strong gains on harder reasoning subsets. In protein, molecular generation and DNA design, the effect is more multi-objective: ordering-aware variants significantly improve selected structural or fragment-constrained metrics while not uniformly dominating the host baseline on every quality metric.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the online bucketized controller tracks the exact DPRM score at empirical-Bernstein rates and that tractable optimization assumptions hold to deliver sample-complexity advantage over random and confidence-only ordering.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DPRM introduces a Doob h-transform process reward module as a plug-in for token ordering in diffusion language models, with convergence proofs and empirical gains over confidence baselines especially on hard reasoning and scientific design tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"DPRM introduces a plug-in module that shifts token ordering in diffusion language models from confidence rules to Doob h-transform process reward guidance.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"8bdfce2e6b9985791c5c4583a9f00ec0138e6b23b391ff5da20c406da8fd670b"},"source":{"id":"2604.24357","kind":"arxiv","version":2},"verdict":{"id":"7dcb89f5-76c2-4738-988f-538c44569db7","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-08T04:09:31.083964Z","strongest_claim":"DPRM improves over confidence-based baselines in pretraining, post-training, test-time scaling, and single-cell masked diffusion, with particularly strong gains on harder reasoning subsets. In protein, molecular generation and DNA design, the effect is more multi-objective: ordering-aware variants significantly improve selected structural or fragment-constrained metrics while not uniformly dominating the host baseline on every quality metric.","one_line_summary":"DPRM introduces a Doob h-transform process reward module as a plug-in for token ordering in diffusion language models, with convergence proofs and empirical gains over confidence baselines especially on hard reasoning and scientific design tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the online bucketized controller tracks the exact DPRM score at empirical-Bernstein rates and that tractable optimization assumptions hold to deliver sample-complexity advantage over random and confidence-only ordering.","pith_extraction_headline":"DPRM introduces a plug-in module that shifts token ordering in diffusion language models from confidence rules to Doob h-transform process reward guidance."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.24357/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-21T06:42:00.638736Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T22:14:02.501214Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"7a145bc4fe3da201ba25241462a0d3e175f8d5c2eddb830d7f625367efce8c8c"},"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":"2604.24357","created_at":"2026-06-19T16:09:58.559457+00:00"},{"alias_kind":"arxiv_version","alias_value":"2604.24357v2","created_at":"2026-06-19T16:09:58.559457+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.24357","created_at":"2026-06-19T16:09:58.559457+00:00"},{"alias_kind":"pith_short_12","alias_value":"LWK7VBFA7CBQ","created_at":"2026-06-19T16:09:58.559457+00:00"},{"alias_kind":"pith_short_16","alias_value":"LWK7VBFA7CBQDKVK","created_at":"2026-06-19T16:09:58.559457+00:00"},{"alias_kind":"pith_short_8","alias_value":"LWK7VBFA","created_at":"2026-06-19T16:09:58.559457+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/LWK7VBFA7CBQDKVKVILWGL6A57","json":"https://pith.science/pith/LWK7VBFA7CBQDKVKVILWGL6A57.json","graph_json":"https://pith.science/api/pith-number/LWK7VBFA7CBQDKVKVILWGL6A57/graph.json","events_json":"https://pith.science/api/pith-number/LWK7VBFA7CBQDKVKVILWGL6A57/events.json","paper":"https://pith.science/paper/LWK7VBFA"},"agent_actions":{"view_html":"https://pith.science/pith/LWK7VBFA7CBQDKVKVILWGL6A57","download_json":"https://pith.science/pith/LWK7VBFA7CBQDKVKVILWGL6A57.json","view_paper":"https://pith.science/paper/LWK7VBFA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2604.24357&json=true","fetch_graph":"https://pith.science/api/pith-number/LWK7VBFA7CBQDKVKVILWGL6A57/graph.json","fetch_events":"https://pith.science/api/pith-number/LWK7VBFA7CBQDKVKVILWGL6A57/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LWK7VBFA7CBQDKVKVILWGL6A57/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LWK7VBFA7CBQDKVKVILWGL6A57/action/storage_attestation","attest_author":"https://pith.science/pith/LWK7VBFA7CBQDKVKVILWGL6A57/action/author_attestation","sign_citation":"https://pith.science/pith/LWK7VBFA7CBQDKVKVILWGL6A57/action/citation_signature","submit_replication":"https://pith.science/pith/LWK7VBFA7CBQDKVKVILWGL6A57/action/replication_record"}},"created_at":"2026-06-19T16:09:58.559457+00:00","updated_at":"2026-06-19T16:09:58.559457+00:00"}