{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:VQNMFQ6AMLSXEZ7MVMNKZT356O","short_pith_number":"pith:VQNMFQ6A","schema_version":"1.0","canonical_sha256":"ac1ac2c3c062e57267ecab1aaccf7df3928d90aa50772ab35230d1fc5702f69b","source":{"kind":"arxiv","id":"2511.16309","version":2},"attestation_state":"computed","paper":{"title":"Sparse Autoencoders are Topic Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Leander Girrbach, Zeynep Akata","submitted_at":"2025-11-20T12:37:54Z","abstract_excerpt":"Sparse autoencoders (SAEs) are used to analyze embeddings, but their role and practical value are debated. We propose a new perspective on SAEs by demonstrating that they can be naturally understood as topic models. We propose a continuous topic model (CTM) inspired by Latent Dirichlet Allocation (LDA) for embedding spaces and derive the SAE objective as a maximum a posteriori estimator under this model. This view implies SAE features are thematic components rather than steerable directions. To confirm our theoretical findings, we introduce SAE-TM, a topic modeling framework that: (1) trains 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":"2511.16309","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-11-20T12:37:54Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"b81fa04cf4346eb3945e90691ad49d162a073575b4fa9ae4af1f3bba9036c7a6","abstract_canon_sha256":"c805adf1dc522d2110f058776bd322493c0afb33b17d7519c8e928689b145017"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:04:17.559667Z","signature_b64":"LFRlSEeiv8y9xnNhpGCToa1UKHvwiBVvRJ2HimQgFVV5D54A8IBcHql2Y5N2pzXuVzdQHe2+DA7w5H8bZZ+/Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ac1ac2c3c062e57267ecab1aaccf7df3928d90aa50772ab35230d1fc5702f69b","last_reissued_at":"2026-05-20T00:04:17.558757Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:04:17.558757Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Sparse Autoencoders are Topic Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Leander Girrbach, Zeynep Akata","submitted_at":"2025-11-20T12:37:54Z","abstract_excerpt":"Sparse autoencoders (SAEs) are used to analyze embeddings, but their role and practical value are debated. We propose a new perspective on SAEs by demonstrating that they can be naturally understood as topic models. We propose a continuous topic model (CTM) inspired by Latent Dirichlet Allocation (LDA) for embedding spaces and derive the SAE objective as a maximum a posteriori estimator under this model. This view implies SAE features are thematic components rather than steerable directions. To confirm our theoretical findings, we introduce SAE-TM, a topic modeling framework that: (1) trains a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2511.16309","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/2511.16309/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":"2511.16309","created_at":"2026-05-20T00:04:17.558911+00:00"},{"alias_kind":"arxiv_version","alias_value":"2511.16309v2","created_at":"2026-05-20T00:04:17.558911+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2511.16309","created_at":"2026-05-20T00:04:17.558911+00:00"},{"alias_kind":"pith_short_12","alias_value":"VQNMFQ6AMLSX","created_at":"2026-05-20T00:04:17.558911+00:00"},{"alias_kind":"pith_short_16","alias_value":"VQNMFQ6AMLSXEZ7M","created_at":"2026-05-20T00:04:17.558911+00:00"},{"alias_kind":"pith_short_8","alias_value":"VQNMFQ6A","created_at":"2026-05-20T00:04:17.558911+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/VQNMFQ6AMLSXEZ7MVMNKZT356O","json":"https://pith.science/pith/VQNMFQ6AMLSXEZ7MVMNKZT356O.json","graph_json":"https://pith.science/api/pith-number/VQNMFQ6AMLSXEZ7MVMNKZT356O/graph.json","events_json":"https://pith.science/api/pith-number/VQNMFQ6AMLSXEZ7MVMNKZT356O/events.json","paper":"https://pith.science/paper/VQNMFQ6A"},"agent_actions":{"view_html":"https://pith.science/pith/VQNMFQ6AMLSXEZ7MVMNKZT356O","download_json":"https://pith.science/pith/VQNMFQ6AMLSXEZ7MVMNKZT356O.json","view_paper":"https://pith.science/paper/VQNMFQ6A","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2511.16309&json=true","fetch_graph":"https://pith.science/api/pith-number/VQNMFQ6AMLSXEZ7MVMNKZT356O/graph.json","fetch_events":"https://pith.science/api/pith-number/VQNMFQ6AMLSXEZ7MVMNKZT356O/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VQNMFQ6AMLSXEZ7MVMNKZT356O/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VQNMFQ6AMLSXEZ7MVMNKZT356O/action/storage_attestation","attest_author":"https://pith.science/pith/VQNMFQ6AMLSXEZ7MVMNKZT356O/action/author_attestation","sign_citation":"https://pith.science/pith/VQNMFQ6AMLSXEZ7MVMNKZT356O/action/citation_signature","submit_replication":"https://pith.science/pith/VQNMFQ6AMLSXEZ7MVMNKZT356O/action/replication_record"}},"created_at":"2026-05-20T00:04:17.558911+00:00","updated_at":"2026-05-20T00:04:17.558911+00:00"}