{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:AWP4YKGS72BW3GXMW4LNZSIXJ3","short_pith_number":"pith:AWP4YKGS","schema_version":"1.0","canonical_sha256":"059fcc28d2fe836d9aecb716dcc9174ecd3eed40b3a918822a2cb013d0b41f92","source":{"kind":"arxiv","id":"2410.01244","version":2},"attestation_state":"computed","paper":{"title":"Robustness and Structure Preservation in Flow-Based Generative Models via Wasserstein Path-Space Divergences","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Equivariant vector fields enable score-based generative models to learn group-invariant distributions without data augmentation.","cross_cats":["cs.LG","math.PR"],"primary_cat":"stat.ML","authors_text":"Benjamin J. Zhang, Markos A. Katsoulakis, Ziyu Chen","submitted_at":"2024-10-02T05:14:28Z","abstract_excerpt":"We introduce a novel Wasserstein-1 ($W_1$) path-space divergence for stochastic and deterministic dynamics and establish a Wasserstein Uncertainty Propagation (WUP) theorem that bounds the $W_1$ distance between terminal distributions by the proposed divergence, equivalently characterized by a weighted $L^2$ discrepancy between the underlying drifts and the $W_1$ distance between their initial measures. A key ingredient is a probabilistic framework combining adjoint Feynman-Kac representations with synchronous coupling (and reflection coupling on bounded domains), yielding Wasserstein stabilit"},"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":"2410.01244","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2024-10-02T05:14:28Z","cross_cats_sorted":["cs.LG","math.PR"],"title_canon_sha256":"27d556e01d8b157f9908df21ad78bd166c06bcf13b9cc3feca19063bc210afd2","abstract_canon_sha256":"c3e76177c1ad4fb5620334875c4f8b76877e9f0d120abfebce4e269ab89f436c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-30T01:17:20.568339Z","signature_b64":"CmxxnZ9GQy2sYSAPeo+JhdHg6OIOYUnaJ0PNBJsAl6DU4Mu9FCZoIc47R9Iz7oct1BfC6mKuTzXyODbicc74Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"059fcc28d2fe836d9aecb716dcc9174ecd3eed40b3a918822a2cb013d0b41f92","last_reissued_at":"2026-06-30T01:17:20.567791Z","signature_status":"signed_v1","first_computed_at":"2026-06-30T01:17:20.567791Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Robustness and Structure Preservation in Flow-Based Generative Models via Wasserstein Path-Space Divergences","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Equivariant vector fields enable score-based generative models to learn group-invariant distributions without data augmentation.","cross_cats":["cs.LG","math.PR"],"primary_cat":"stat.ML","authors_text":"Benjamin J. Zhang, Markos A. Katsoulakis, Ziyu Chen","submitted_at":"2024-10-02T05:14:28Z","abstract_excerpt":"We introduce a novel Wasserstein-1 ($W_1$) path-space divergence for stochastic and deterministic dynamics and establish a Wasserstein Uncertainty Propagation (WUP) theorem that bounds the $W_1$ distance between terminal distributions by the proposed divergence, equivalently characterized by a weighted $L^2$ discrepancy between the underlying drifts and the $W_1$ distance between their initial measures. A key ingredient is a probabilistic framework combining adjoint Feynman-Kac representations with synchronous coupling (and reflection coupling on bounded domains), yielding Wasserstein stabilit"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"one can learn the score of a symmetrized distribution using equivariant vector fields without data augmentations through the analysis of the optimality and equivalence of score-matching objectives. This also provides practical guidance that one does not have to augment the dataset as long as the vector field or the neural network parametrization is equivariant.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The data distribution is exactly group-invariant and the symmetry group is known in advance so that an exactly equivariant vector field can be constructed; the improved d1 bound and the HJB equivalence both rest on this premise (abstract, paragraphs on improved d1 bound and HJB analysis).","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Equivariant SGMs achieve improved Wasserstein-1 generalization bounds on group-invariant distributions and learn the symmetrized score via equivariant vector fields without augmentation, with non-equivariant models incurring a quantifiable model-form error.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Equivariant vector fields enable score-based generative models to learn group-invariant distributions without data augmentation.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"19435ceb3f2896af078439d40ac2ef8cdc9a1aa65f748ef88ed5d0e7273c95c7"},"source":{"id":"2410.01244","kind":"arxiv","version":2},"verdict":{"id":"ea2aa5ea-9787-4cf2-887f-a63dfdbb2fb5","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-23T20:31:36.267911Z","strongest_claim":"one can learn the score of a symmetrized distribution using equivariant vector fields without data augmentations through the analysis of the optimality and equivalence of score-matching objectives. This also provides practical guidance that one does not have to augment the dataset as long as the vector field or the neural network parametrization is equivariant.","one_line_summary":"Equivariant SGMs achieve improved Wasserstein-1 generalization bounds on group-invariant distributions and learn the symmetrized score via equivariant vector fields without augmentation, with non-equivariant models incurring a quantifiable model-form error.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The data distribution is exactly group-invariant and the symmetry group is known in advance so that an exactly equivariant vector field can be constructed; the improved d1 bound and the HJB equivalence both rest on this premise (abstract, paragraphs on improved d1 bound and HJB analysis).","pith_extraction_headline":"Equivariant vector fields enable score-based generative models to learn group-invariant distributions without data augmentation."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2410.01244/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":"2410.01244","created_at":"2026-06-30T01:17:20.567864+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.01244v2","created_at":"2026-06-30T01:17:20.567864+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.01244","created_at":"2026-06-30T01:17:20.567864+00:00"},{"alias_kind":"pith_short_12","alias_value":"AWP4YKGS72BW","created_at":"2026-06-30T01:17:20.567864+00:00"},{"alias_kind":"pith_short_16","alias_value":"AWP4YKGS72BW3GXM","created_at":"2026-06-30T01:17:20.567864+00:00"},{"alias_kind":"pith_short_8","alias_value":"AWP4YKGS","created_at":"2026-06-30T01:17:20.567864+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.06140","citing_title":"SymDrift: One-Shot Generative Modeling under Symmetries","ref_index":46,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AWP4YKGS72BW3GXMW4LNZSIXJ3","json":"https://pith.science/pith/AWP4YKGS72BW3GXMW4LNZSIXJ3.json","graph_json":"https://pith.science/api/pith-number/AWP4YKGS72BW3GXMW4LNZSIXJ3/graph.json","events_json":"https://pith.science/api/pith-number/AWP4YKGS72BW3GXMW4LNZSIXJ3/events.json","paper":"https://pith.science/paper/AWP4YKGS"},"agent_actions":{"view_html":"https://pith.science/pith/AWP4YKGS72BW3GXMW4LNZSIXJ3","download_json":"https://pith.science/pith/AWP4YKGS72BW3GXMW4LNZSIXJ3.json","view_paper":"https://pith.science/paper/AWP4YKGS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.01244&json=true","fetch_graph":"https://pith.science/api/pith-number/AWP4YKGS72BW3GXMW4LNZSIXJ3/graph.json","fetch_events":"https://pith.science/api/pith-number/AWP4YKGS72BW3GXMW4LNZSIXJ3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AWP4YKGS72BW3GXMW4LNZSIXJ3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AWP4YKGS72BW3GXMW4LNZSIXJ3/action/storage_attestation","attest_author":"https://pith.science/pith/AWP4YKGS72BW3GXMW4LNZSIXJ3/action/author_attestation","sign_citation":"https://pith.science/pith/AWP4YKGS72BW3GXMW4LNZSIXJ3/action/citation_signature","submit_replication":"https://pith.science/pith/AWP4YKGS72BW3GXMW4LNZSIXJ3/action/replication_record"}},"created_at":"2026-06-30T01:17:20.567864+00:00","updated_at":"2026-06-30T01:17:20.567864+00:00"}