{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:23VNEKNIXMWRCND4CTRNMVTPYH","short_pith_number":"pith:23VNEKNI","schema_version":"1.0","canonical_sha256":"d6ead229a8bb2d11347c14e2d6566fc1c12d1e0bd9c6d5a6a19b8e09f2007d13","source":{"kind":"arxiv","id":"2410.08201","version":2},"attestation_state":"computed","paper":{"title":"Efficient Dictionary Learning with Switch Sparse Autoencoders","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Anish Mudide, Christian Schroeder de Witt, Eric J. Michaud, Joshua Engels, Max Tegmark","submitted_at":"2024-10-10T17:59:11Z","abstract_excerpt":"Sparse autoencoders (SAEs) are a recent technique for decomposing neural network activations into human-interpretable features. However, in order for SAEs to identify all features represented in frontier models, it will be necessary to scale them up to very high width, posing a computational challenge. In this work, we introduce Switch Sparse Autoencoders, a novel SAE architecture aimed at reducing the compute cost of training SAEs. Inspired by sparse mixture of experts models, Switch SAEs route activation vectors between smaller \"expert\" SAEs, enabling SAEs to efficiently scale to many more f"},"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":"2410.08201","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-10-10T17:59:11Z","cross_cats_sorted":[],"title_canon_sha256":"5724e03f7c64b3a140d0e52b8cd5c99b8ba27e6f0344e706b821aec3ab3f8d37","abstract_canon_sha256":"949a906e32082035f73aa9e761e948c26cf0051f6c746371dd08e262d202a0d7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:14:24.990447Z","signature_b64":"EbY7vE/mSAM2ddqwNzgSoImlgvl/JFH5crtZLb3kQjzD3D9OcgBEIo+1+QYGfhXr1cPP3709AODMq5kqtfbaBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d6ead229a8bb2d11347c14e2d6566fc1c12d1e0bd9c6d5a6a19b8e09f2007d13","last_reissued_at":"2026-07-05T11:14:24.989941Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:14:24.989941Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Efficient Dictionary Learning with Switch Sparse Autoencoders","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Anish Mudide, Christian Schroeder de Witt, Eric J. Michaud, Joshua Engels, Max Tegmark","submitted_at":"2024-10-10T17:59:11Z","abstract_excerpt":"Sparse autoencoders (SAEs) are a recent technique for decomposing neural network activations into human-interpretable features. However, in order for SAEs to identify all features represented in frontier models, it will be necessary to scale them up to very high width, posing a computational challenge. In this work, we introduce Switch Sparse Autoencoders, a novel SAE architecture aimed at reducing the compute cost of training SAEs. Inspired by sparse mixture of experts models, Switch SAEs route activation vectors between smaller \"expert\" SAEs, enabling SAEs to efficiently scale to many more f"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.08201","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/2410.08201/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":"2410.08201","created_at":"2026-07-05T11:14:24.990005+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.08201v2","created_at":"2026-07-05T11:14:24.990005+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.08201","created_at":"2026-07-05T11:14:24.990005+00:00"},{"alias_kind":"pith_short_12","alias_value":"23VNEKNIXMWR","created_at":"2026-07-05T11:14:24.990005+00:00"},{"alias_kind":"pith_short_16","alias_value":"23VNEKNIXMWRCND4","created_at":"2026-07-05T11:14:24.990005+00:00"},{"alias_kind":"pith_short_8","alias_value":"23VNEKNI","created_at":"2026-07-05T11:14:24.990005+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.01799","citing_title":"Expander Sparse Autoencoders: Parameter-Efficient Dictionaries for Mechanistic Interpretability","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07714","citing_title":"Beyond Accuracy: Interpreting Topic Representation in Suicide Ideation Detection Models","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28149","citing_title":"Sign-Aware Gated Sparse Autoencoders: Modeling Anticorrelated Features with Bi-Jump-ReLU Activations","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2601.14004","citing_title":"Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models","ref_index":213,"is_internal_anchor":false},{"citing_arxiv_id":"2604.04946","citing_title":"Sparse Autoencoders as a Steering Basis for Phase Synchronization in Graph-Based CFD Surrogates","ref_index":8,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/23VNEKNIXMWRCND4CTRNMVTPYH","json":"https://pith.science/pith/23VNEKNIXMWRCND4CTRNMVTPYH.json","graph_json":"https://pith.science/api/pith-number/23VNEKNIXMWRCND4CTRNMVTPYH/graph.json","events_json":"https://pith.science/api/pith-number/23VNEKNIXMWRCND4CTRNMVTPYH/events.json","paper":"https://pith.science/paper/23VNEKNI"},"agent_actions":{"view_html":"https://pith.science/pith/23VNEKNIXMWRCND4CTRNMVTPYH","download_json":"https://pith.science/pith/23VNEKNIXMWRCND4CTRNMVTPYH.json","view_paper":"https://pith.science/paper/23VNEKNI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.08201&json=true","fetch_graph":"https://pith.science/api/pith-number/23VNEKNIXMWRCND4CTRNMVTPYH/graph.json","fetch_events":"https://pith.science/api/pith-number/23VNEKNIXMWRCND4CTRNMVTPYH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/23VNEKNIXMWRCND4CTRNMVTPYH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/23VNEKNIXMWRCND4CTRNMVTPYH/action/storage_attestation","attest_author":"https://pith.science/pith/23VNEKNIXMWRCND4CTRNMVTPYH/action/author_attestation","sign_citation":"https://pith.science/pith/23VNEKNIXMWRCND4CTRNMVTPYH/action/citation_signature","submit_replication":"https://pith.science/pith/23VNEKNIXMWRCND4CTRNMVTPYH/action/replication_record"}},"created_at":"2026-07-05T11:14:24.990005+00:00","updated_at":"2026-07-05T11:14:24.990005+00:00"}