{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:BC3FLE7LBA4UNZ4ZQECK3UTXLL","short_pith_number":"pith:BC3FLE7L","schema_version":"1.0","canonical_sha256":"08b65593eb083946e7998104add2775af2df471220874ef85699208bc5aa091b","source":{"kind":"arxiv","id":"2410.08417","version":2},"attestation_state":"computed","paper":{"title":"Bilinear MLPs enable weight-based mechanistic interpretability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Alice Rigg, Jose M. Oramas, Lee Sharkey, Michael T. Pearce, Thomas Dooms","submitted_at":"2024-10-10T23:22:11Z","abstract_excerpt":"A mechanistic understanding of how MLPs do computation in deep neural networks remains elusive. Current interpretability work can extract features from hidden activations over an input dataset but generally cannot explain how MLP weights construct features. One challenge is that element-wise nonlinearities introduce higher-order interactions and make it difficult to trace computations through the MLP layer. In this paper, we analyze bilinear MLPs, a type of Gated Linear Unit (GLU) without any element-wise nonlinearity that nevertheless achieves competitive performance. Bilinear MLPs can be ful"},"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":"2410.08417","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-10-10T23:22:11Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"f0219b66dcdad6003806f4ed2e406a3997a40be6332395ed08d8e52379f13334","abstract_canon_sha256":"5475ff2cc91ed2fabf9bac81abf84dfb02549b474e9f0897d972421a517fb286"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:26:44.303848Z","signature_b64":"deRbVQ6T939iG1X0/l2RNMUwdwBO+usFY9RJojxMemvbsd0Jj2enU4aJ2cdH5jJcQYBh8SAB4mAmY6AhAiPfCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"08b65593eb083946e7998104add2775af2df471220874ef85699208bc5aa091b","last_reissued_at":"2026-07-05T11:26:44.303339Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:26:44.303339Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Bilinear MLPs enable weight-based mechanistic interpretability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Alice Rigg, Jose M. Oramas, Lee Sharkey, Michael T. Pearce, Thomas Dooms","submitted_at":"2024-10-10T23:22:11Z","abstract_excerpt":"A mechanistic understanding of how MLPs do computation in deep neural networks remains elusive. Current interpretability work can extract features from hidden activations over an input dataset but generally cannot explain how MLP weights construct features. One challenge is that element-wise nonlinearities introduce higher-order interactions and make it difficult to trace computations through the MLP layer. In this paper, we analyze bilinear MLPs, a type of Gated Linear Unit (GLU) without any element-wise nonlinearity that nevertheless achieves competitive performance. Bilinear MLPs can be ful"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.08417","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/2410.08417/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":"2410.08417","created_at":"2026-07-05T11:26:44.303400+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.08417v2","created_at":"2026-07-05T11:26:44.303400+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.08417","created_at":"2026-07-05T11:26:44.303400+00:00"},{"alias_kind":"pith_short_12","alias_value":"BC3FLE7LBA4U","created_at":"2026-07-05T11:26:44.303400+00:00"},{"alias_kind":"pith_short_16","alias_value":"BC3FLE7LBA4UNZ4Z","created_at":"2026-07-05T11:26:44.303400+00:00"},{"alias_kind":"pith_short_8","alias_value":"BC3FLE7L","created_at":"2026-07-05T11:26:44.303400+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.12770","citing_title":"WriteSAE: Sparse Autoencoders for Recurrent State","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2508.05463","citing_title":"Task complexity shapes internal representations and robustness in neural networks","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12770","citing_title":"WriteSAE: Sparse Autoencoders for Recurrent State","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15183","citing_title":"When Are Two Networks the Same? Tensor Similarity for Mechanistic Interpretability","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12770","citing_title":"WriteSAE: Sparse Autoencoders for Recurrent State","ref_index":7,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BC3FLE7LBA4UNZ4ZQECK3UTXLL","json":"https://pith.science/pith/BC3FLE7LBA4UNZ4ZQECK3UTXLL.json","graph_json":"https://pith.science/api/pith-number/BC3FLE7LBA4UNZ4ZQECK3UTXLL/graph.json","events_json":"https://pith.science/api/pith-number/BC3FLE7LBA4UNZ4ZQECK3UTXLL/events.json","paper":"https://pith.science/paper/BC3FLE7L"},"agent_actions":{"view_html":"https://pith.science/pith/BC3FLE7LBA4UNZ4ZQECK3UTXLL","download_json":"https://pith.science/pith/BC3FLE7LBA4UNZ4ZQECK3UTXLL.json","view_paper":"https://pith.science/paper/BC3FLE7L","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.08417&json=true","fetch_graph":"https://pith.science/api/pith-number/BC3FLE7LBA4UNZ4ZQECK3UTXLL/graph.json","fetch_events":"https://pith.science/api/pith-number/BC3FLE7LBA4UNZ4ZQECK3UTXLL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BC3FLE7LBA4UNZ4ZQECK3UTXLL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BC3FLE7LBA4UNZ4ZQECK3UTXLL/action/storage_attestation","attest_author":"https://pith.science/pith/BC3FLE7LBA4UNZ4ZQECK3UTXLL/action/author_attestation","sign_citation":"https://pith.science/pith/BC3FLE7LBA4UNZ4ZQECK3UTXLL/action/citation_signature","submit_replication":"https://pith.science/pith/BC3FLE7LBA4UNZ4ZQECK3UTXLL/action/replication_record"}},"created_at":"2026-07-05T11:26:44.303400+00:00","updated_at":"2026-07-05T11:26:44.303400+00:00"}