{"paper":{"title":"Simple Self-Conditioning Adaptation for Masked Diffusion Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A post-training adaptation conditions masked diffusion models on their own prior clean predictions to enable repeated refinement across denoising steps.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Ferdinando Fioretto, Huu Binh Ta, Michael Cardei","submitted_at":"2026-04-28T19:34:04Z","abstract_excerpt":"Masked diffusion models (MDMs) generate discrete sequences by iterative denoising under an absorbing masking process. In standard masked diffusion, if a token remains masked after a reverse update, the model discards its clean-state prediction for that position. Thus, still-masked positions must be repeatedly inferred from the mask token alone. This design choice limits cross-step refinement. To address this limitation, this paper proposes a simple, yet effective, post-training adaptation for MDMs that conditions each denoising step on the model's own previous clean-state predictions. The resu"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"SCMDM achieves nearly a 50% reduction in generative perplexity on OWT-trained models (42.89 to 23.72), alongside strong improvements in discretized image synthesis quality, small molecular generation, and enhanced fidelity in genomic distribution modeling.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"Once the model's self-generated clean-state estimates become informative, specialization to refinement is preferable to mixing conditional and unconditional objectives in the post-training regime.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SCMDM adapts trained masked diffusion models to condition denoising steps on their own prior clean predictions, cutting generative perplexity nearly in half on open-web text while improving discretized image, molecule, and genomic synthesis.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A post-training adaptation conditions masked diffusion models on their own prior clean predictions to enable repeated refinement across denoising steps.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"1031e3ca8e3a7f6998a969bfdea338dd0f272c0e0d9ddbd8a555a8bbf2969d9f"},"source":{"id":"2604.26985","kind":"arxiv","version":2},"verdict":{"id":"5d41238d-17d8-46ba-b50c-ce4165a35d59","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-07T16:32:57.468001Z","strongest_claim":"SCMDM achieves nearly a 50% reduction in generative perplexity on OWT-trained models (42.89 to 23.72), alongside strong improvements in discretized image synthesis quality, small molecular generation, and enhanced fidelity in genomic distribution modeling.","one_line_summary":"SCMDM adapts trained masked diffusion models to condition denoising steps on their own prior clean predictions, cutting generative perplexity nearly in half on open-web text while improving discretized image, molecule, and genomic synthesis.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"Once the model's self-generated clean-state estimates become informative, specialization to refinement is preferable to mixing conditional and unconditional objectives in the post-training regime.","pith_extraction_headline":"A post-training adaptation conditions masked diffusion models on their own prior clean predictions to enable repeated refinement across denoising steps."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.26985/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-21T03:35:58.343813Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T20:34:09.529655Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"6c95415a457d647223be2874b32f89b25ad77f7ff86651fd48d783b4597d70e2"},"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"}