{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:GIUZVWIM4IABHHPE7XSFMXYNB5","short_pith_number":"pith:GIUZVWIM","schema_version":"1.0","canonical_sha256":"32299ad90ce200139de4fde4565f0d0f4fa144f09e9263b926afe15033732910","source":{"kind":"arxiv","id":"2502.12892","version":2},"attestation_state":"computed","paper":{"title":"Archetypal SAE: Adaptive and Stable Dictionary Learning for Concept Extraction in Large Vision Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Binxu Wang, Demba Ba, Ekdeep Singh Lubana, Isabel Papadimitriou, Jacob S. Prince, Martin Wattenberg, Matthew Kowal, Talia Konkle, Thomas Fel, Victor Boutin","submitted_at":"2025-02-18T14:29:11Z","abstract_excerpt":"Sparse Autoencoders (SAEs) have emerged as a powerful framework for machine learning interpretability, enabling the unsupervised decomposition of model representations into a dictionary of abstract, human-interpretable concepts. However, we reveal a fundamental limitation: existing SAEs exhibit severe instability, as identical models trained on similar datasets can produce sharply different dictionaries, undermining their reliability as an interpretability tool. To address this issue, we draw inspiration from the Archetypal Analysis framework introduced by Cutler & Breiman (1994) and present A"},"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":"2502.12892","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-02-18T14:29:11Z","cross_cats_sorted":[],"title_canon_sha256":"074234fef2259dd47de46c4c3b259630d82b4482587b6c25fa45c9abf8dac652","abstract_canon_sha256":"1d40228afdf2ee289cec8bed75b3db5ba5a20ec9bc3299ad47dadc68b154b356"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:08:40.379821Z","signature_b64":"0swggkKwpsqrOyxoVL6IgnX3kZwPBIBehlFSs24v3dTfpF3CkgUt7I9E2Ap6nh+67Ar6hoTfvOqtYdQbIEKLCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"32299ad90ce200139de4fde4565f0d0f4fa144f09e9263b926afe15033732910","last_reissued_at":"2026-07-05T11:08:40.379306Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:08:40.379306Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Archetypal SAE: Adaptive and Stable Dictionary Learning for Concept Extraction in Large Vision Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Binxu Wang, Demba Ba, Ekdeep Singh Lubana, Isabel Papadimitriou, Jacob S. Prince, Martin Wattenberg, Matthew Kowal, Talia Konkle, Thomas Fel, Victor Boutin","submitted_at":"2025-02-18T14:29:11Z","abstract_excerpt":"Sparse Autoencoders (SAEs) have emerged as a powerful framework for machine learning interpretability, enabling the unsupervised decomposition of model representations into a dictionary of abstract, human-interpretable concepts. However, we reveal a fundamental limitation: existing SAEs exhibit severe instability, as identical models trained on similar datasets can produce sharply different dictionaries, undermining their reliability as an interpretability tool. To address this issue, we draw inspiration from the Archetypal Analysis framework introduced by Cutler & Breiman (1994) and present A"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.12892","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/2502.12892/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":"2502.12892","created_at":"2026-07-05T11:08:40.379378+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.12892v2","created_at":"2026-07-05T11:08:40.379378+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.12892","created_at":"2026-07-05T11:08:40.379378+00:00"},{"alias_kind":"pith_short_12","alias_value":"GIUZVWIM4IAB","created_at":"2026-07-05T11:08:40.379378+00:00"},{"alias_kind":"pith_short_16","alias_value":"GIUZVWIM4IABHHPE","created_at":"2026-07-05T11:08:40.379378+00:00"},{"alias_kind":"pith_short_8","alias_value":"GIUZVWIM","created_at":"2026-07-05T11:08:40.379378+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.18667","citing_title":"Can neurons speak? Semantic narration of vision at single-cell resolution","ref_index":30,"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":33,"is_internal_anchor":false},{"citing_arxiv_id":"2509.03738","citing_title":"Mechanistic Interpretability with Sparse Autoencoder Neural Operators","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00474","citing_title":"From Local to Global to Mechanistic: An iERF-Centered Unified Framework for Interpreting Vision Models","ref_index":36,"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":13,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GIUZVWIM4IABHHPE7XSFMXYNB5","json":"https://pith.science/pith/GIUZVWIM4IABHHPE7XSFMXYNB5.json","graph_json":"https://pith.science/api/pith-number/GIUZVWIM4IABHHPE7XSFMXYNB5/graph.json","events_json":"https://pith.science/api/pith-number/GIUZVWIM4IABHHPE7XSFMXYNB5/events.json","paper":"https://pith.science/paper/GIUZVWIM"},"agent_actions":{"view_html":"https://pith.science/pith/GIUZVWIM4IABHHPE7XSFMXYNB5","download_json":"https://pith.science/pith/GIUZVWIM4IABHHPE7XSFMXYNB5.json","view_paper":"https://pith.science/paper/GIUZVWIM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.12892&json=true","fetch_graph":"https://pith.science/api/pith-number/GIUZVWIM4IABHHPE7XSFMXYNB5/graph.json","fetch_events":"https://pith.science/api/pith-number/GIUZVWIM4IABHHPE7XSFMXYNB5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GIUZVWIM4IABHHPE7XSFMXYNB5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GIUZVWIM4IABHHPE7XSFMXYNB5/action/storage_attestation","attest_author":"https://pith.science/pith/GIUZVWIM4IABHHPE7XSFMXYNB5/action/author_attestation","sign_citation":"https://pith.science/pith/GIUZVWIM4IABHHPE7XSFMXYNB5/action/citation_signature","submit_replication":"https://pith.science/pith/GIUZVWIM4IABHHPE7XSFMXYNB5/action/replication_record"}},"created_at":"2026-07-05T11:08:40.379378+00:00","updated_at":"2026-07-05T11:08:40.379378+00:00"}