{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:SVPP5BZ4KRL3U3BYV6Z66AFFC5","short_pith_number":"pith:SVPP5BZ4","schema_version":"1.0","canonical_sha256":"955efe873c5457ba6c38afb3ef00a51763176a1b821fb80f8acd7229dcc2aebc","source":{"kind":"arxiv","id":"2106.13314","version":1},"attestation_state":"computed","paper":{"title":"Promises and Pitfalls of Black-Box Concept Learning Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Anita Mahinpei, Finale Doshi-Velez, Isaac Lage, Justin Clark, Weiwei Pan","submitted_at":"2021-06-24T21:00:28Z","abstract_excerpt":"Machine learning models that incorporate concept learning as an intermediate step in their decision making process can match the performance of black-box predictive models while retaining the ability to explain outcomes in human understandable terms. However, we demonstrate that the concept representations learned by these models encode information beyond the pre-defined concepts, and that natural mitigation strategies do not fully work, rendering the interpretation of the downstream prediction misleading. We describe the mechanism underlying the information leakage and suggest recourse for mi"},"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":"2106.13314","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2021-06-24T21:00:28Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"561167027eb5df0b0e3b15939d6c854a35173a5f3a6b85bb8d05dcbca79e2337","abstract_canon_sha256":"34fbe49407ba0f529a37d53190fff7bcb4c09ee228e8c6040bf093ef997c8725"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:52:18.048126Z","signature_b64":"G3cgCf8T+WsFQ2dwoZO0QJOtYMdgtXqv5if8yG6oywz9vEIgdYxeKRjTW/MapQP5sER/91S6yqvNPOn9RQ8LDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"955efe873c5457ba6c38afb3ef00a51763176a1b821fb80f8acd7229dcc2aebc","last_reissued_at":"2026-07-05T02:52:18.047594Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:52:18.047594Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Promises and Pitfalls of Black-Box Concept Learning Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Anita Mahinpei, Finale Doshi-Velez, Isaac Lage, Justin Clark, Weiwei Pan","submitted_at":"2021-06-24T21:00:28Z","abstract_excerpt":"Machine learning models that incorporate concept learning as an intermediate step in their decision making process can match the performance of black-box predictive models while retaining the ability to explain outcomes in human understandable terms. However, we demonstrate that the concept representations learned by these models encode information beyond the pre-defined concepts, and that natural mitigation strategies do not fully work, rendering the interpretation of the downstream prediction misleading. We describe the mechanism underlying the information leakage and suggest recourse for mi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2106.13314","kind":"arxiv","version":1},"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/2106.13314/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":"2106.13314","created_at":"2026-07-05T02:52:18.047654+00:00"},{"alias_kind":"arxiv_version","alias_value":"2106.13314v1","created_at":"2026-07-05T02:52:18.047654+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2106.13314","created_at":"2026-07-05T02:52:18.047654+00:00"},{"alias_kind":"pith_short_12","alias_value":"SVPP5BZ4KRL3","created_at":"2026-07-05T02:52:18.047654+00:00"},{"alias_kind":"pith_short_16","alias_value":"SVPP5BZ4KRL3U3BY","created_at":"2026-07-05T02:52:18.047654+00:00"},{"alias_kind":"pith_short_8","alias_value":"SVPP5BZ4","created_at":"2026-07-05T02:52:18.047654+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.19489","citing_title":"Concept Flow Models: Anchoring Concept-Based Reasoning with Hierarchical Bottlenecks","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2606.10669","citing_title":"In Defense of Information Leakage in Concept-based Models","ref_index":185,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00578","citing_title":"Caption Bottleneck Models","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04364","citing_title":"Spatially Grounded Concept Bottleneck Models via Part-Factorized Attention","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04326","citing_title":"Measuring What Matters: Synthetic Benchmarks for Concept Bottleneck Models","ref_index":53,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30498","citing_title":"On the Faithfulness of Post-Hoc Concept Bottleneck Models","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29836","citing_title":"CB-SLICE: Concept-Based Interpretable Error Slice Discovery","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16076","citing_title":"Prototype-Grounded Concept Models for Verifiable Concept Alignment","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20693","citing_title":"Interpretable Discriminative Text Representations via Agreement and Label Disentanglement","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20908","citing_title":"SynCB: A Synergy Concept-Based Model with Dynamic Routing Between Concepts and Complementary Neural Branches","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08482","citing_title":"ShifaMind: A Multiplicative Concept Bottleneck for Interpretable ICD-10 Coding","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07140","citing_title":"Neurosymbolic Framework for Concept-Driven Logical Reasoning in Skeleton-Based Human Action Recognition","ref_index":73,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16042","citing_title":"Towards Intrinsic Interpretability of Large Language Models:A Survey of Design Principles and Architectures","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16076","citing_title":"Prototype-Grounded Concept Models for Verifiable Concept Alignment","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19323","citing_title":"Concept Inconsistency in Dermoscopic Concept Bottleneck Models: A Rough-Set Analysis of the Derm7pt Dataset","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SVPP5BZ4KRL3U3BYV6Z66AFFC5","json":"https://pith.science/pith/SVPP5BZ4KRL3U3BYV6Z66AFFC5.json","graph_json":"https://pith.science/api/pith-number/SVPP5BZ4KRL3U3BYV6Z66AFFC5/graph.json","events_json":"https://pith.science/api/pith-number/SVPP5BZ4KRL3U3BYV6Z66AFFC5/events.json","paper":"https://pith.science/paper/SVPP5BZ4"},"agent_actions":{"view_html":"https://pith.science/pith/SVPP5BZ4KRL3U3BYV6Z66AFFC5","download_json":"https://pith.science/pith/SVPP5BZ4KRL3U3BYV6Z66AFFC5.json","view_paper":"https://pith.science/paper/SVPP5BZ4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2106.13314&json=true","fetch_graph":"https://pith.science/api/pith-number/SVPP5BZ4KRL3U3BYV6Z66AFFC5/graph.json","fetch_events":"https://pith.science/api/pith-number/SVPP5BZ4KRL3U3BYV6Z66AFFC5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SVPP5BZ4KRL3U3BYV6Z66AFFC5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SVPP5BZ4KRL3U3BYV6Z66AFFC5/action/storage_attestation","attest_author":"https://pith.science/pith/SVPP5BZ4KRL3U3BYV6Z66AFFC5/action/author_attestation","sign_citation":"https://pith.science/pith/SVPP5BZ4KRL3U3BYV6Z66AFFC5/action/citation_signature","submit_replication":"https://pith.science/pith/SVPP5BZ4KRL3U3BYV6Z66AFFC5/action/replication_record"}},"created_at":"2026-07-05T02:52:18.047654+00:00","updated_at":"2026-07-05T02:52:18.047654+00:00"}