{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:ER37LBHXQI2KKKSFXRFB625FJT","short_pith_number":"pith:ER37LBHX","schema_version":"1.0","canonical_sha256":"2477f584f78234a52a45bc4a1f6ba54cfa9be6e1c2f06e19793a1867f9fade0d","source":{"kind":"arxiv","id":"2305.03452","version":1},"attestation_state":"computed","paper":{"title":"A technical note on bilinear layers for interpretability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.NE"],"primary_cat":"cs.LG","authors_text":"Lee Sharkey","submitted_at":"2023-05-05T11:56:26Z","abstract_excerpt":"The ability of neural networks to represent more features than neurons makes interpreting them challenging. This phenomenon, known as superposition, has spurred efforts to find architectures that are more interpretable than standard multilayer perceptrons (MLPs) with elementwise activation functions. In this note, I examine bilinear layers, which are a type of MLP layer that are mathematically much easier to analyze while simultaneously performing better than standard MLPs. Although they are nonlinear functions of their input, I demonstrate that bilinear layers can be expressed using only line"},"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":"2305.03452","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-05-05T11:56:26Z","cross_cats_sorted":["cs.NE"],"title_canon_sha256":"5e26f446cfd5d3221760f8833cbc2d558350aed8046a45e5ea3344ad92da3f9e","abstract_canon_sha256":"db72b5dafb32fa46808ea72acedbe5af19f77252882778b8829d3ad5e90a2a26"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:07:23.374052Z","signature_b64":"A/awiCau1ll81PjbX+uhkhZgFrCFXHCy+wVZL3Qr21cvISVE3MyAv9TFQiEEhPGDCdNtXLm9S3w7oSeSZaYlAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2477f584f78234a52a45bc4a1f6ba54cfa9be6e1c2f06e19793a1867f9fade0d","last_reissued_at":"2026-07-05T06:07:23.373646Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:07:23.373646Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A technical note on bilinear layers for interpretability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.NE"],"primary_cat":"cs.LG","authors_text":"Lee Sharkey","submitted_at":"2023-05-05T11:56:26Z","abstract_excerpt":"The ability of neural networks to represent more features than neurons makes interpreting them challenging. This phenomenon, known as superposition, has spurred efforts to find architectures that are more interpretable than standard multilayer perceptrons (MLPs) with elementwise activation functions. In this note, I examine bilinear layers, which are a type of MLP layer that are mathematically much easier to analyze while simultaneously performing better than standard MLPs. Although they are nonlinear functions of their input, I demonstrate that bilinear layers can be expressed using only line"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2305.03452","kind":"arxiv","version":1},"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/2305.03452/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":"2305.03452","created_at":"2026-07-05T06:07:23.373705+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.03452v1","created_at":"2026-07-05T06:07:23.373705+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.03452","created_at":"2026-07-05T06:07:23.373705+00:00"},{"alias_kind":"pith_short_12","alias_value":"ER37LBHXQI2K","created_at":"2026-07-05T06:07:23.373705+00:00"},{"alias_kind":"pith_short_16","alias_value":"ER37LBHXQI2KKKSF","created_at":"2026-07-05T06:07:23.373705+00:00"},{"alias_kind":"pith_short_8","alias_value":"ER37LBHX","created_at":"2026-07-05T06:07:23.373705+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.07414","citing_title":"Sparsely gated tiny linear experts","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2606.02385","citing_title":"How Optimality Structures Sparse Dictionaries: A Theory for Understanding SAE Representations","ref_index":221,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15183","citing_title":"When Are Two Networks the Same? Tensor Similarity for Mechanistic Interpretability","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12809","citing_title":"Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces","ref_index":151,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ER37LBHXQI2KKKSFXRFB625FJT","json":"https://pith.science/pith/ER37LBHXQI2KKKSFXRFB625FJT.json","graph_json":"https://pith.science/api/pith-number/ER37LBHXQI2KKKSFXRFB625FJT/graph.json","events_json":"https://pith.science/api/pith-number/ER37LBHXQI2KKKSFXRFB625FJT/events.json","paper":"https://pith.science/paper/ER37LBHX"},"agent_actions":{"view_html":"https://pith.science/pith/ER37LBHXQI2KKKSFXRFB625FJT","download_json":"https://pith.science/pith/ER37LBHXQI2KKKSFXRFB625FJT.json","view_paper":"https://pith.science/paper/ER37LBHX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.03452&json=true","fetch_graph":"https://pith.science/api/pith-number/ER37LBHXQI2KKKSFXRFB625FJT/graph.json","fetch_events":"https://pith.science/api/pith-number/ER37LBHXQI2KKKSFXRFB625FJT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ER37LBHXQI2KKKSFXRFB625FJT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ER37LBHXQI2KKKSFXRFB625FJT/action/storage_attestation","attest_author":"https://pith.science/pith/ER37LBHXQI2KKKSFXRFB625FJT/action/author_attestation","sign_citation":"https://pith.science/pith/ER37LBHXQI2KKKSFXRFB625FJT/action/citation_signature","submit_replication":"https://pith.science/pith/ER37LBHXQI2KKKSFXRFB625FJT/action/replication_record"}},"created_at":"2026-07-05T06:07:23.373705+00:00","updated_at":"2026-07-05T06:07:23.373705+00:00"}