{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:PDCYIKFFMHIKRG23HVUTGXJV3D","short_pith_number":"pith:PDCYIKFF","schema_version":"1.0","canonical_sha256":"78c58428a561d0a89b5b3d69335d35d8e850b4f9d70b8d7dbd4e1dfe105e7b04","source":{"kind":"arxiv","id":"2412.05276","version":2},"attestation_state":"computed","paper":{"title":"Sparse autoencoders reveal selective remapping of visual concepts during adaptation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Hyesu Lim, Jaegul Choo, Jinho Choi, Steffen Schneider","submitted_at":"2024-12-06T18:59:51Z","abstract_excerpt":"Adapting foundation models for specific purposes has become a standard approach to build machine learning systems for downstream applications. Yet, it is an open question which mechanisms take place during adaptation. Here we develop a new Sparse Autoencoder (SAE) for the CLIP vision transformer, named PatchSAE, to extract interpretable concepts at granular levels (e.g., shape, color, or semantics of an object) and their patch-wise spatial attributions. We explore how these concepts influence the model output in downstream image classification tasks and investigate how recent state-of-the-art "},"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":"2412.05276","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2024-12-06T18:59:51Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"3f17586abca0d1cc8a399a29b3fcd68946598bf6f02371b382e0be576131a73e","abstract_canon_sha256":"023ba691df7465dbe59baff2687d6fc2b8c3281ddee7d51fdf816e0d9a765874"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:36:32.536739Z","signature_b64":"4xXsiW5vQMuZ7/GTf961Q5vcHmHConUwPn5nR6YnM1yFHjVNEaAzmDpoIk8tkwMtKQ3yriPkcljYvdfbq9ddDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"78c58428a561d0a89b5b3d69335d35d8e850b4f9d70b8d7dbd4e1dfe105e7b04","last_reissued_at":"2026-07-05T10:36:32.535976Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:36:32.535976Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Sparse autoencoders reveal selective remapping of visual concepts during adaptation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Hyesu Lim, Jaegul Choo, Jinho Choi, Steffen Schneider","submitted_at":"2024-12-06T18:59:51Z","abstract_excerpt":"Adapting foundation models for specific purposes has become a standard approach to build machine learning systems for downstream applications. Yet, it is an open question which mechanisms take place during adaptation. Here we develop a new Sparse Autoencoder (SAE) for the CLIP vision transformer, named PatchSAE, to extract interpretable concepts at granular levels (e.g., shape, color, or semantics of an object) and their patch-wise spatial attributions. We explore how these concepts influence the model output in downstream image classification tasks and investigate how recent state-of-the-art "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.05276","kind":"arxiv","version":2},"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/2412.05276/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":"2412.05276","created_at":"2026-07-05T10:36:32.536076+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.05276v2","created_at":"2026-07-05T10:36:32.536076+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.05276","created_at":"2026-07-05T10:36:32.536076+00:00"},{"alias_kind":"pith_short_12","alias_value":"PDCYIKFFMHIK","created_at":"2026-07-05T10:36:32.536076+00:00"},{"alias_kind":"pith_short_16","alias_value":"PDCYIKFFMHIKRG23","created_at":"2026-07-05T10:36:32.536076+00:00"},{"alias_kind":"pith_short_8","alias_value":"PDCYIKFF","created_at":"2026-07-05T10:36:32.536076+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21197","citing_title":"Extraction and Analysis of Multimodal Concepts in Vision Language Models through Sparse Autoencoders","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2506.01247","citing_title":"Beyond Interpretability: When, Why, and How Sparse Autoencoders Enable Label-Free Visual Steering","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01829","citing_title":"GeoSAE: Geometric Prior-Guided Layer-Wise Sparse Autoencoder Annotation of Brain MRI Foundation Models","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13304","citing_title":"Can Cross-Layer Transcoders Replace Vision Transformer Activations? An Interpretable Perspective on Vision","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07802","citing_title":"Latent Anomaly Knowledge Excavation: Unveiling Sparse Sensitive Neurons in Vision-Language Models","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2604.05724","citing_title":"Beyond Semantics: Disentangling Information Scope in Sparse Autoencoders for CLIP","ref_index":20,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PDCYIKFFMHIKRG23HVUTGXJV3D","json":"https://pith.science/pith/PDCYIKFFMHIKRG23HVUTGXJV3D.json","graph_json":"https://pith.science/api/pith-number/PDCYIKFFMHIKRG23HVUTGXJV3D/graph.json","events_json":"https://pith.science/api/pith-number/PDCYIKFFMHIKRG23HVUTGXJV3D/events.json","paper":"https://pith.science/paper/PDCYIKFF"},"agent_actions":{"view_html":"https://pith.science/pith/PDCYIKFFMHIKRG23HVUTGXJV3D","download_json":"https://pith.science/pith/PDCYIKFFMHIKRG23HVUTGXJV3D.json","view_paper":"https://pith.science/paper/PDCYIKFF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.05276&json=true","fetch_graph":"https://pith.science/api/pith-number/PDCYIKFFMHIKRG23HVUTGXJV3D/graph.json","fetch_events":"https://pith.science/api/pith-number/PDCYIKFFMHIKRG23HVUTGXJV3D/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PDCYIKFFMHIKRG23HVUTGXJV3D/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PDCYIKFFMHIKRG23HVUTGXJV3D/action/storage_attestation","attest_author":"https://pith.science/pith/PDCYIKFFMHIKRG23HVUTGXJV3D/action/author_attestation","sign_citation":"https://pith.science/pith/PDCYIKFFMHIKRG23HVUTGXJV3D/action/citation_signature","submit_replication":"https://pith.science/pith/PDCYIKFFMHIKRG23HVUTGXJV3D/action/replication_record"}},"created_at":"2026-07-05T10:36:32.536076+00:00","updated_at":"2026-07-05T10:36:32.536076+00:00"}