{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:D7PZPQKJQM4X73VXPJVISRJ7C5","short_pith_number":"pith:D7PZPQKJ","schema_version":"1.0","canonical_sha256":"1fdf97c14983397feeb77a6a89453f175b2feda1b46697093df366f08ab43660","source":{"kind":"arxiv","id":"2603.04198","version":2},"attestation_state":"computed","paper":{"title":"Stable and Steerable Sparse Autoencoders with Weight Regularization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Oliver M. Crook, Piotr Jedryszek","submitted_at":"2026-03-04T15:46:23Z","abstract_excerpt":"Sparse autoencoders (SAEs) are widely used to extract human-interpretable features from neural network activations, but their learned features can vary substantially across random seeds and training choices. To improve stability, we studied weight regularization by adding L1 or L2 penalties on encoder and decoder weights, and evaluate how regularization interacts with common SAE training defaults. On MNIST, we observe that L2 weight regularization produces a core of highly aligned features and, when combined with tied initialization and unit-norm decoder constraints, it dramatically increases "},"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":"2603.04198","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2026-03-04T15:46:23Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"8a1df9e08fd141c7e5936ec03ca4688a60a6e405bc58c44cf41b39da1218a33e","abstract_canon_sha256":"a72b8ad555a29a0d57c631b5fcb60d2c723496a2865151813293f8a47d402866"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-19T16:10:36.213635Z","signature_b64":"b6gwMXvTDkKUfuVOuH8L5dttZeBdQhGUUxAzOCl8KqyVwWgxluR/0ocj3XHG+Wj28f/27WK8efx/s06+HYLLAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1fdf97c14983397feeb77a6a89453f175b2feda1b46697093df366f08ab43660","last_reissued_at":"2026-06-19T16:10:36.213124Z","signature_status":"signed_v1","first_computed_at":"2026-06-19T16:10:36.213124Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Stable and Steerable Sparse Autoencoders with Weight Regularization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Oliver M. Crook, Piotr Jedryszek","submitted_at":"2026-03-04T15:46:23Z","abstract_excerpt":"Sparse autoencoders (SAEs) are widely used to extract human-interpretable features from neural network activations, but their learned features can vary substantially across random seeds and training choices. To improve stability, we studied weight regularization by adding L1 or L2 penalties on encoder and decoder weights, and evaluate how regularization interacts with common SAE training defaults. On MNIST, we observe that L2 weight regularization produces a core of highly aligned features and, when combined with tied initialization and unit-norm decoder constraints, it dramatically increases "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.04198","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/2603.04198/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":"2603.04198","created_at":"2026-06-19T16:10:36.213187+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.04198v2","created_at":"2026-06-19T16:10:36.213187+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.04198","created_at":"2026-06-19T16:10:36.213187+00:00"},{"alias_kind":"pith_short_12","alias_value":"D7PZPQKJQM4X","created_at":"2026-06-19T16:10:36.213187+00:00"},{"alias_kind":"pith_short_16","alias_value":"D7PZPQKJQM4X73VX","created_at":"2026-06-19T16:10:36.213187+00:00"},{"alias_kind":"pith_short_8","alias_value":"D7PZPQKJ","created_at":"2026-06-19T16:10:36.213187+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/D7PZPQKJQM4X73VXPJVISRJ7C5","json":"https://pith.science/pith/D7PZPQKJQM4X73VXPJVISRJ7C5.json","graph_json":"https://pith.science/api/pith-number/D7PZPQKJQM4X73VXPJVISRJ7C5/graph.json","events_json":"https://pith.science/api/pith-number/D7PZPQKJQM4X73VXPJVISRJ7C5/events.json","paper":"https://pith.science/paper/D7PZPQKJ"},"agent_actions":{"view_html":"https://pith.science/pith/D7PZPQKJQM4X73VXPJVISRJ7C5","download_json":"https://pith.science/pith/D7PZPQKJQM4X73VXPJVISRJ7C5.json","view_paper":"https://pith.science/paper/D7PZPQKJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.04198&json=true","fetch_graph":"https://pith.science/api/pith-number/D7PZPQKJQM4X73VXPJVISRJ7C5/graph.json","fetch_events":"https://pith.science/api/pith-number/D7PZPQKJQM4X73VXPJVISRJ7C5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/D7PZPQKJQM4X73VXPJVISRJ7C5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/D7PZPQKJQM4X73VXPJVISRJ7C5/action/storage_attestation","attest_author":"https://pith.science/pith/D7PZPQKJQM4X73VXPJVISRJ7C5/action/author_attestation","sign_citation":"https://pith.science/pith/D7PZPQKJQM4X73VXPJVISRJ7C5/action/citation_signature","submit_replication":"https://pith.science/pith/D7PZPQKJQM4X73VXPJVISRJ7C5/action/replication_record"}},"created_at":"2026-06-19T16:10:36.213187+00:00","updated_at":"2026-06-19T16:10:36.213187+00:00"}