{"paper":{"title":"StAD: Stein Amortized Divergence for Fast Likelihoods with Diffusion and Flow","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"StAD uses the Langevin-Stein operator to learn PF-ODE divergences without ever computing Jacobians.","cross_cats":["astro-ph.IM","cs.LG"],"primary_cat":"stat.ML","authors_text":"Gurjeet Jagwani, Hiranya Peiris, Sinan Deger, Stephen Thorp","submitted_at":"2026-05-15T18:00:00Z","abstract_excerpt":"Diffusion and flow-based models are ubiquitously used for generative modelling and density estimation. They admit a deterministic probability flow ordinary differential equation (PF-ODE), analogous to continuous normalizing flows (CNFs), which describes the transport of the probability mass. Obtaining the likelihood from these models is of interest to many workflows, especially Bayesian analysis, and requires solving the trace of the Jacobian to compute the divergence of the learned PF-ODE, which is either $\\mathcal{O}(D^2)$ to compute exactly or $\\mathcal{O}(D)$ with a noisy estimate. We intr"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We introduce StAD, a new distillation method to predict and learn the divergence of the PF-ODE using the Langevin-Stein operator without ever computing the Jacobian.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"Under some regularity conditions these learned vector fields can be made to satisfy the Stein class, enabling generalization across a varied class of generative models.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"StAD distills divergence of PF-ODEs via the Langevin-Stein operator for faster, lower-variance likelihood estimation in generative models without Jacobian costs.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"StAD uses the Langevin-Stein operator to learn PF-ODE divergences without ever computing Jacobians.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b667f0ee796635ea1db8bd9597046a8c827e97ec753575321684eb98ff11c12e"},"source":{"id":"2605.16486","kind":"arxiv","version":1},"verdict":{"id":"0eaf6edf-d2a7-46e3-99a2-c6734be8b5da","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T21:29:40.354359Z","strongest_claim":"We introduce StAD, a new distillation method to predict and learn the divergence of the PF-ODE using the Langevin-Stein operator without ever computing the Jacobian.","one_line_summary":"StAD distills divergence of PF-ODEs via the Langevin-Stein operator for faster, lower-variance likelihood estimation in generative models without Jacobian costs.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"Under some regularity conditions these learned vector fields can be made to satisfy the Stein class, enabling generalization across a varied class of generative models.","pith_extraction_headline":"StAD uses the Langevin-Stein operator to learn PF-ODE divergences without ever computing Jacobians."},"integrity":{"clean":false,"summary":{"advisory":4,"critical":0,"by_detector":{"doi_compliance":{"total":4,"advisory":4,"critical":0,"informational":0}},"informational":0},"endpoint":"/pith/2605.16486/integrity.json","findings":[{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.1093/mnras/stz1960.Alsing) was visible in the surrounding text but could not be confirmed against doi.org as printed.","detector":"doi_compliance","severity":"advisory","ref_index":1,"audited_at":"2026-05-19T21:41:21.806016Z","detected_doi":"10.1093/mnras/stz1960.Alsing","finding_type":"recoverable_identifier","verdict_class":"incontrovertible","detected_arxiv_id":null},{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.3847/1538-3881/ad54bf.10) was visible in the surrounding text but could not be confirmed against doi.org as printed.","detector":"doi_compliance","severity":"advisory","ref_index":5,"audited_at":"2026-05-19T21:41:21.806016Z","detected_doi":"10.3847/1538-3881/ad54bf.10","finding_type":"recoverable_identifier","verdict_class":"incontrovertible","detected_arxiv_id":null},{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.1609/aaai.v32i1.11671.13) was visible in the surrounding text but could not be confirmed against doi.org as printed.","detector":"doi_compliance","severity":"advisory","ref_index":18,"audited_at":"2026-05-19T21:41:21.806016Z","detected_doi":"10.1609/aaai.v32i1.11671.13","finding_type":"recoverable_identifier","verdict_class":"incontrovertible","detected_arxiv_id":null},{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.1007/978-1-4612-0919-5_26.Ronneberger) was visible in the surrounding text but could not be confirmed against doi.org as printed.","detector":"doi_compliance","severity":"advisory","ref_index":21,"audited_at":"2026-05-19T21:41:21.806016Z","detected_doi":"10.1007/978-1-4612-0919-5_26.Ronneberger","finding_type":"recoverable_identifier","verdict_class":"incontrovertible","detected_arxiv_id":null}],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T22:01:23.173257Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T21:41:21.806016Z","status":"completed","version":"1.0.0","findings_count":4},{"name":"ai_meta_artifact","ran_at":"2026-05-19T19:33:23.109541Z","status":"skipped","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T19:21:57.027751Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"ecbae6ba49def4b58d207d095156e44478f5ddd5e28e8c4dde70de3b648fce2c"},"references":{"count":29,"sample":[{"doi":"10.1093/mnras/","year":2023,"title":"Monthly Notices of the Royal Astronomical Society428(4), 3121–3138 (2013) https://doi.org/10.1093/mnras/ sts261","work_id":"7288268a-13f4-4979-8480-3f091b17e16a","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"cc/paper_files/paper/2019/file/ ba7609ee5789cc4dff171045a693a65f- Paper.pdf","work_id":"cfd1ce49-968a-4991-9300-9b1eabc44e4a","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"cc/paper_files/paper/2023/file/ 9a8eb202c060b7d81f5889631cbcd47e- Paper-Conference.pdf","work_id":"a54abfd4-3f37-44ad-b4d4-6d6786c7c900","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"A downsampled variant of imagenet as an alternative to the CIFAR datasets","work_id":"b28da76e-30a0-4103-a956-55ca47f3e243","ref_index":4,"cited_arxiv_id":"1707.08819","is_internal_anchor":true},{"doi":"10.3847/1538-","year":2015,"title":"NICE: Non-linear Independent Components Estimation","work_id":"47bfe556-ce63-4e62-b810-488f772bcb82","ref_index":5,"cited_arxiv_id":"1410.8516","is_internal_anchor":true}],"resolved_work":29,"snapshot_sha256":"160f63e437f41e5394a1dbb0f3b5b3b861cb621ce70b1195c2545de9a361bb39","internal_anchors":4},"formal_canon":{"evidence_count":2,"snapshot_sha256":"e3d52a5a3402d7732bf2ac497f0229e82ad009efcba6d14e38140ea32c4cbe1f"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}