{"paper":{"title":"From Baselines to Transport Geodesics: Axiomatic Attribution via Optimal Generative Flows","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Aumann-Shapley line integrals along transport geodesics give unique and stable attributions.","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Cenwei Zhang, Lei You, Lin Zhu, Manxi Lin","submitted_at":"2026-03-05T12:05:20Z","abstract_excerpt":"Feature attributions often hide a critical modeling choice: they explain a prediction along a counterfactual path from a reference state to an input. Different baselines, interpolations, and generative trajectories define different paths and can therefor produce different explanations. We study this path ambiguity as a modeling problem. Our central question is whether the path can be chosen by the data-generating transport process, rather than by a hand-designed interpolation or by the sensitivity geometry of the model being explained. We separate attribution into fixed-path credit allocation "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"For a fixed path, the Aumann-Shapley line integral is the unique attribution rule under standard fixed-path axioms and explicit coordinate-trace regularity. For path selection, minimizing kinetic action over flows that transport a reference distribution to the data distribution yields a transport-geodesic attribution principle.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the data-generating process can be adequately modeled by flows whose kinetic action minimization produces attribution paths that are meaningfully better than hand-designed interpolations.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Transport-geodesic attribution via optimal generative flows selects principled paths for feature attributions by minimizing kinetic action.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Aumann-Shapley line integrals along transport geodesics give unique and stable attributions.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b71dbf6bf5820f53d38bbc81fdc7a6e174ccd7d7d7992c4d211cec849bc40aa8"},"source":{"id":"2603.05093","kind":"arxiv","version":2},"verdict":{"id":"e797a979-c1fe-4c5d-ac1c-0245a58b5ede","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T15:59:44.986282Z","strongest_claim":"For a fixed path, the Aumann-Shapley line integral is the unique attribution rule under standard fixed-path axioms and explicit coordinate-trace regularity. For path selection, minimizing kinetic action over flows that transport a reference distribution to the data distribution yields a transport-geodesic attribution principle.","one_line_summary":"Transport-geodesic attribution via optimal generative flows selects principled paths for feature attributions by minimizing kinetic action.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the data-generating process can be adequately modeled by flows whose kinetic action minimization produces attribution paths that are meaningfully better than hand-designed interpolations.","pith_extraction_headline":"Aumann-Shapley line integrals along transport geodesics give unique and stable attributions."},"references":{"count":11,"sample":[{"doi":"","year":2006,"title":"Shapley explainability on the data manifold","work_id":"35a2b27f-26fb-4426-addc-cd80e8066b93","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Progressive Growing of GANs for Improved Quality, Stability, and Variation","work_id":"5e8c5f57-fe56-4018-ac90-b6b3b849f420","ref_index":3,"cited_arxiv_id":"1710.10196","is_internal_anchor":true},{"doi":"","year":null,"title":"Flow Matching for Generative Modeling","work_id":"6edb71c4-5d64-40af-a394-9757ea051a36","ref_index":4,"cited_arxiv_id":"2210.02747","is_internal_anchor":true},{"doi":"","year":null,"title":"Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow","work_id":"a1989e1b-d66d-4533-be3a-fb9c5fd62290","ref_index":5,"cited_arxiv_id":"2209.03003","is_internal_anchor":true},{"doi":"","year":2017,"title":"URL https://proceedings.neurips","work_id":"a90e4413-c221-491a-9870-2dcd060cc979","ref_index":6,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":11,"snapshot_sha256":"14dcd104cc939b34262256480f2d8ad11402a8b7973122c7160aab61d5e9f9f6","internal_anchors":6},"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"}