{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:DKJ22TW5JC73RK7BUZWQO6GLDT","short_pith_number":"pith:DKJ22TW5","canonical_record":{"source":{"id":"2605.14210","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T00:08:09Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"b0ae81c6429213f37b091cc4ee8ff3711c5735dd2bee6ef6ba2603a15cd2313c","abstract_canon_sha256":"d18b36420cce8c01aefc0ac9e31af333976332de468b499340243316eea4c4cb"},"schema_version":"1.0"},"canonical_sha256":"1a93ad4edd48bfb8abe1a66d0778cb1cdc24ff0f720300a8d2fe56ebfb2c8d01","source":{"kind":"arxiv","id":"2605.14210","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14210","created_at":"2026-05-17T23:39:10Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14210v1","created_at":"2026-05-17T23:39:10Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14210","created_at":"2026-05-17T23:39:10Z"},{"alias_kind":"pith_short_12","alias_value":"DKJ22TW5JC73","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"DKJ22TW5JC73RK7B","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"DKJ22TW5","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:DKJ22TW5JC73RK7BUZWQO6GLDT","target":"record","payload":{"canonical_record":{"source":{"id":"2605.14210","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T00:08:09Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"b0ae81c6429213f37b091cc4ee8ff3711c5735dd2bee6ef6ba2603a15cd2313c","abstract_canon_sha256":"d18b36420cce8c01aefc0ac9e31af333976332de468b499340243316eea4c4cb"},"schema_version":"1.0"},"canonical_sha256":"1a93ad4edd48bfb8abe1a66d0778cb1cdc24ff0f720300a8d2fe56ebfb2c8d01","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:10.938768Z","signature_b64":"rkTOTCj5BQpOFUllPFsyiqn0HEgpsZijvLh+O99cfSEkMi0Cs+ZrUSq5b/P1YyPyVWylVnWZGG+6E2MURv6eAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1a93ad4edd48bfb8abe1a66d0778cb1cdc24ff0f720300a8d2fe56ebfb2c8d01","last_reissued_at":"2026-05-17T23:39:10.938204Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:10.938204Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.14210","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:10Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"1Xpt9xwDiNvcZzXYSuYsfKEgtoWBm+o67QESp25i41Sag6vvks8PItrwCsOZzbl8ybfV3Eg85rxYtjBfNSEbAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-25T23:29:30.522361Z"},"content_sha256":"d65f9554f778753ab83420c6560a4c9ef82426df5eee39e14269fc1093e1bd6a","schema_version":"1.0","event_id":"sha256:d65f9554f778753ab83420c6560a4c9ef82426df5eee39e14269fc1093e1bd6a"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:DKJ22TW5JC73RK7BUZWQO6GLDT","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Towards Fine-Grained and Verifiable Concept Bottleneck Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A verifiable CBM framework grounds concepts in localized image patches, achieving comparable accuracy to standard CBMs on medical benchmarks while enabling direct inspection of concept correctness.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Guang Yang, Haijie Xu, Mariathasan Anish, Shuang Wu, Yingying Fang","submitted_at":"2026-05-14T00:08:09Z","abstract_excerpt":"Concept Bottleneck Models (CBMs) offer interpretable alternatives to black-box predictors by introducing human-relatable concepts before the final output. However, existing CBMs struggle to verify whether predicted concepts correspond to the correct visual evidence, limiting their reliability. We propose a fine-grained CBM framework that grounds each concept in localized visual evidence, enabling direct inspection of where and how concepts are encoded. This design allows users to interpret predictions and verify that the model learns intended concepts rather than spurious correlations. Experim"},"claims":{"count":3,"items":[{"kind":"strongest_claim","text":"Experiments on medical imaging benchmarks show that our learned concept space is information-complete and achieves predictive performance comparable to standard CBMs, while substantially improving transparency. Unlike post-hoc attribution methods, our framework validates both the presence and correctness of concept representations.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That localizing each concept to visual evidence regions will reliably prevent the model from learning spurious correlations and that human inspection of these regions will correctly verify intended concepts without additional validation data or metrics.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A verifiable CBM framework grounds concepts in localized image patches, achieving comparable accuracy to standard CBMs on medical benchmarks while enabling direct inspection of concept correctness.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"}],"snapshot_sha256":"3d8ec82b864ceafb7784f1d2682f53ef52d338a193f1e9df4d43a77aa554c43c"},"source":{"id":"2605.14210","kind":"arxiv","version":1},"verdict":{"id":"96eb17ae-6b1a-42cc-98d6-4e7b63ce7db9","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:11:05.265582Z","strongest_claim":"Experiments on medical imaging benchmarks show that our learned concept space is information-complete and achieves predictive performance comparable to standard CBMs, while substantially improving transparency. Unlike post-hoc attribution methods, our framework validates both the presence and correctness of concept representations.","one_line_summary":"A verifiable CBM framework grounds concepts in localized image patches, achieving comparable accuracy to standard CBMs on medical benchmarks while enabling direct inspection of concept correctness.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That localizing each concept to visual evidence regions will reliably prevent the model from learning spurious correlations and that human inspection of these regions will correctly verify intended concepts without additional validation data or metrics.","pith_extraction_headline":""},"references":{"count":28,"sample":[{"doi":"","year":2021,"title":"Nature Machine Intelligence3(12), 1061–1070 (2021)","work_id":"0e86d0b4-4b5b-481e-8b03-bd784549f5b8","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Nature Communications15(1), 524 (2024)","work_id":"5581de68-cfa3-4689-9666-1f84c3d6808c","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"In: 2018 IEEE winter conference on applications of computer vision (WACV)","work_id":"0225a284-acca-44fe-9981-76614cc172c2","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Skin lesion analysis toward melanoma detection 2018: A challenge hosted by the international skin imaging collaboration (isic)","work_id":"a1680554-54cc-449f-aa92-0acb940cad0a","ref_index":4,"cited_arxiv_id":"1902.03368","is_internal_anchor":true},{"doi":"","year":2022,"title":"Advances in neural information processing systems35, 21400–21413 (2022)","work_id":"8353caaa-c464-446c-947b-09005ed0c9fd","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":28,"snapshot_sha256":"f533aa522abb6dcac79e29e35b3b584cc061455c59c70f00c8b531f597c877e4","internal_anchors":2},"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"},"verdict_id":"96eb17ae-6b1a-42cc-98d6-4e7b63ce7db9"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:10Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"EQfvR0Ksla0WuWyFn0S6qJu19Nd23ATtYvMBJk0mcydwazAEtED04nUtPhzUZCUoRi5AijDuyMntDbSmuUYDDQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-25T23:29:30.523611Z"},"content_sha256":"b836833f29bdc676f375f67e0a5b4d0190ffd731b4d5e7ff4dfc1923bacda898","schema_version":"1.0","event_id":"sha256:b836833f29bdc676f375f67e0a5b4d0190ffd731b4d5e7ff4dfc1923bacda898"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/DKJ22TW5JC73RK7BUZWQO6GLDT/bundle.json","state_url":"https://pith.science/pith/DKJ22TW5JC73RK7BUZWQO6GLDT/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/DKJ22TW5JC73RK7BUZWQO6GLDT/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-25T23:29:30Z","links":{"resolver":"https://pith.science/pith/DKJ22TW5JC73RK7BUZWQO6GLDT","bundle":"https://pith.science/pith/DKJ22TW5JC73RK7BUZWQO6GLDT/bundle.json","state":"https://pith.science/pith/DKJ22TW5JC73RK7BUZWQO6GLDT/state.json","well_known_bundle":"https://pith.science/.well-known/pith/DKJ22TW5JC73RK7BUZWQO6GLDT/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:DKJ22TW5JC73RK7BUZWQO6GLDT","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"d18b36420cce8c01aefc0ac9e31af333976332de468b499340243316eea4c4cb","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T00:08:09Z","title_canon_sha256":"b0ae81c6429213f37b091cc4ee8ff3711c5735dd2bee6ef6ba2603a15cd2313c"},"schema_version":"1.0","source":{"id":"2605.14210","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14210","created_at":"2026-05-17T23:39:10Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14210v1","created_at":"2026-05-17T23:39:10Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14210","created_at":"2026-05-17T23:39:10Z"},{"alias_kind":"pith_short_12","alias_value":"DKJ22TW5JC73","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"DKJ22TW5JC73RK7B","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"DKJ22TW5","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:b836833f29bdc676f375f67e0a5b4d0190ffd731b4d5e7ff4dfc1923bacda898","target":"graph","created_at":"2026-05-17T23:39:10Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":3,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Experiments on medical imaging benchmarks show that our learned concept space is information-complete and achieves predictive performance comparable to standard CBMs, while substantially improving transparency. Unlike post-hoc attribution methods, our framework validates both the presence and correctness of concept representations."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That localizing each concept to visual evidence regions will reliably prevent the model from learning spurious correlations and that human inspection of these regions will correctly verify intended concepts without additional validation data or metrics."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"A verifiable CBM framework grounds concepts in localized image patches, achieving comparable accuracy to standard CBMs on medical benchmarks while enabling direct inspection of concept correctness."}],"snapshot_sha256":"3d8ec82b864ceafb7784f1d2682f53ef52d338a193f1e9df4d43a77aa554c43c"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Concept Bottleneck Models (CBMs) offer interpretable alternatives to black-box predictors by introducing human-relatable concepts before the final output. However, existing CBMs struggle to verify whether predicted concepts correspond to the correct visual evidence, limiting their reliability. We propose a fine-grained CBM framework that grounds each concept in localized visual evidence, enabling direct inspection of where and how concepts are encoded. This design allows users to interpret predictions and verify that the model learns intended concepts rather than spurious correlations. Experim","authors_text":"Guang Yang, Haijie Xu, Mariathasan Anish, Shuang Wu, Yingying Fang","cross_cats":["cs.AI"],"headline":"A verifiable CBM framework grounds concepts in localized image patches, achieving comparable accuracy to standard CBMs on medical benchmarks while enabling direct inspection of concept correctness.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T00:08:09Z","title":"Towards Fine-Grained and Verifiable Concept Bottleneck Models"},"references":{"count":28,"internal_anchors":2,"resolved_work":28,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Nature Machine Intelligence3(12), 1061–1070 (2021)","work_id":"0e86d0b4-4b5b-481e-8b03-bd784549f5b8","year":2021},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Nature Communications15(1), 524 (2024)","work_id":"5581de68-cfa3-4689-9666-1f84c3d6808c","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"In: 2018 IEEE winter conference on applications of computer vision (WACV)","work_id":"0225a284-acca-44fe-9981-76614cc172c2","year":2018},{"cited_arxiv_id":"1902.03368","doi":"","is_internal_anchor":true,"ref_index":4,"title":"Skin lesion analysis toward melanoma detection 2018: A challenge hosted by the international skin imaging collaboration (isic)","work_id":"a1680554-54cc-449f-aa92-0acb940cad0a","year":2018},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Advances in neural information processing systems35, 21400–21413 (2022)","work_id":"8353caaa-c464-446c-947b-09005ed0c9fd","year":2022}],"snapshot_sha256":"f533aa522abb6dcac79e29e35b3b584cc061455c59c70f00c8b531f597c877e4"},"source":{"id":"2605.14210","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T02:11:05.265582Z","id":"96eb17ae-6b1a-42cc-98d6-4e7b63ce7db9","model_set":{"reader":"grok-4.3"},"one_line_summary":"A verifiable CBM framework grounds concepts in localized image patches, achieving comparable accuracy to standard CBMs on medical benchmarks while enabling direct inspection of concept correctness.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"","strongest_claim":"Experiments on medical imaging benchmarks show that our learned concept space is information-complete and achieves predictive performance comparable to standard CBMs, while substantially improving transparency. Unlike post-hoc attribution methods, our framework validates both the presence and correctness of concept representations.","weakest_assumption":"That localizing each concept to visual evidence regions will reliably prevent the model from learning spurious correlations and that human inspection of these regions will correctly verify intended concepts without additional validation data or metrics."}},"verdict_id":"96eb17ae-6b1a-42cc-98d6-4e7b63ce7db9"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:d65f9554f778753ab83420c6560a4c9ef82426df5eee39e14269fc1093e1bd6a","target":"record","created_at":"2026-05-17T23:39:10Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"d18b36420cce8c01aefc0ac9e31af333976332de468b499340243316eea4c4cb","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T00:08:09Z","title_canon_sha256":"b0ae81c6429213f37b091cc4ee8ff3711c5735dd2bee6ef6ba2603a15cd2313c"},"schema_version":"1.0","source":{"id":"2605.14210","kind":"arxiv","version":1}},"canonical_sha256":"1a93ad4edd48bfb8abe1a66d0778cb1cdc24ff0f720300a8d2fe56ebfb2c8d01","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"1a93ad4edd48bfb8abe1a66d0778cb1cdc24ff0f720300a8d2fe56ebfb2c8d01","first_computed_at":"2026-05-17T23:39:10.938204Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:10.938204Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"rkTOTCj5BQpOFUllPFsyiqn0HEgpsZijvLh+O99cfSEkMi0Cs+ZrUSq5b/P1YyPyVWylVnWZGG+6E2MURv6eAw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:10.938768Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.14210","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:d65f9554f778753ab83420c6560a4c9ef82426df5eee39e14269fc1093e1bd6a","sha256:b836833f29bdc676f375f67e0a5b4d0190ffd731b4d5e7ff4dfc1923bacda898"],"state_sha256":"721eef0ef0a7c10bb2acb202838b532284dbb1a9bd4d6b247614ff2c6f1cda60"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"ptPpHMAKSYoQPDvGAOwZMe6xO3Ul+UpsUClI9rEGQr3fz2xrTlWW51BTiUx2jzcp5XLXoX9R1/nwjVmoYNm5Bw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-25T23:29:30.528692Z","bundle_sha256":"6769677a38f3426975734ba93e4e8d6535582dd64ae854dc26db899ca5bf12af"}}