{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:PV5PWMLA3ED3A76YGJXGRC562D","short_pith_number":"pith:PV5PWMLA","schema_version":"1.0","canonical_sha256":"7d7afb3160d907b07fd8326e688bbed0ff045c52eab70cfcc388fac2c7293811","source":{"kind":"arxiv","id":"2504.17921","version":3},"attestation_state":"computed","paper":{"title":"Avoiding Leakage Poisoning: Concept Interventions Under Distribution Shifts","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CR","cs.HC"],"primary_cat":"cs.LG","authors_text":"Gabriele Dominici, Mateja Jamnik, Mateo Espinosa Zarlenga, Pietro Barbiero, Zohreh Shams","submitted_at":"2025-04-24T20:24:31Z","abstract_excerpt":"In this paper, we investigate how concept-based models (CMs) respond to out-of-distribution (OOD) inputs. CMs are interpretable neural architectures that first predict a set of high-level concepts (e.g., stripes, black) and then predict a task label from those concepts. In particular, we study the impact of concept interventions (i.e., operations where a human expert corrects a CM's mispredicted concepts at test time) on CMs' task predictions when inputs are OOD. Our analysis reveals a weakness in current state-of-the-art CMs, which we term leakage poisoning, that prevents them from properly i"},"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":"2504.17921","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-04-24T20:24:31Z","cross_cats_sorted":["cs.AI","cs.CR","cs.HC"],"title_canon_sha256":"8c856e18baf9b0d131e9a51442c5781e10e1a923f02ca9aa7560464258078a45","abstract_canon_sha256":"d4b8776bff322d1e9f41f9c13979b33e1bfe4e3f31bf0e1fcc5fc7aeb1cb9790"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:48:02.932046Z","signature_b64":"zImYVH6ZTLif2tBq/tUX0FipoXzh0QBaMS4j8odSECBmCqOeRQOkUv+zHVGwObwTGHisZ79fFvPfMrMrTUsjCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7d7afb3160d907b07fd8326e688bbed0ff045c52eab70cfcc388fac2c7293811","last_reissued_at":"2026-07-05T11:48:02.931559Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:48:02.931559Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Avoiding Leakage Poisoning: Concept Interventions Under Distribution Shifts","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CR","cs.HC"],"primary_cat":"cs.LG","authors_text":"Gabriele Dominici, Mateja Jamnik, Mateo Espinosa Zarlenga, Pietro Barbiero, Zohreh Shams","submitted_at":"2025-04-24T20:24:31Z","abstract_excerpt":"In this paper, we investigate how concept-based models (CMs) respond to out-of-distribution (OOD) inputs. CMs are interpretable neural architectures that first predict a set of high-level concepts (e.g., stripes, black) and then predict a task label from those concepts. In particular, we study the impact of concept interventions (i.e., operations where a human expert corrects a CM's mispredicted concepts at test time) on CMs' task predictions when inputs are OOD. Our analysis reveals a weakness in current state-of-the-art CMs, which we term leakage poisoning, that prevents them from properly i"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.17921","kind":"arxiv","version":3},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2504.17921/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":"2504.17921","created_at":"2026-07-05T11:48:02.931611+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.17921v3","created_at":"2026-07-05T11:48:02.931611+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.17921","created_at":"2026-07-05T11:48:02.931611+00:00"},{"alias_kind":"pith_short_12","alias_value":"PV5PWMLA3ED3","created_at":"2026-07-05T11:48:02.931611+00:00"},{"alias_kind":"pith_short_16","alias_value":"PV5PWMLA3ED3A76Y","created_at":"2026-07-05T11:48:02.931611+00:00"},{"alias_kind":"pith_short_8","alias_value":"PV5PWMLA","created_at":"2026-07-05T11:48:02.931611+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/PV5PWMLA3ED3A76YGJXGRC562D","json":"https://pith.science/pith/PV5PWMLA3ED3A76YGJXGRC562D.json","graph_json":"https://pith.science/api/pith-number/PV5PWMLA3ED3A76YGJXGRC562D/graph.json","events_json":"https://pith.science/api/pith-number/PV5PWMLA3ED3A76YGJXGRC562D/events.json","paper":"https://pith.science/paper/PV5PWMLA"},"agent_actions":{"view_html":"https://pith.science/pith/PV5PWMLA3ED3A76YGJXGRC562D","download_json":"https://pith.science/pith/PV5PWMLA3ED3A76YGJXGRC562D.json","view_paper":"https://pith.science/paper/PV5PWMLA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.17921&json=true","fetch_graph":"https://pith.science/api/pith-number/PV5PWMLA3ED3A76YGJXGRC562D/graph.json","fetch_events":"https://pith.science/api/pith-number/PV5PWMLA3ED3A76YGJXGRC562D/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PV5PWMLA3ED3A76YGJXGRC562D/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PV5PWMLA3ED3A76YGJXGRC562D/action/storage_attestation","attest_author":"https://pith.science/pith/PV5PWMLA3ED3A76YGJXGRC562D/action/author_attestation","sign_citation":"https://pith.science/pith/PV5PWMLA3ED3A76YGJXGRC562D/action/citation_signature","submit_replication":"https://pith.science/pith/PV5PWMLA3ED3A76YGJXGRC562D/action/replication_record"}},"created_at":"2026-07-05T11:48:02.931611+00:00","updated_at":"2026-07-05T11:48:02.931611+00:00"}