{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:A3AMMYM6NSKPZCKSHAS53LU7OU","short_pith_number":"pith:A3AMMYM6","schema_version":"1.0","canonical_sha256":"06c0c6619e6c94fc89523825ddae9f751a85753a3a8b7a62b5406ee457644c3e","source":{"kind":"arxiv","id":"2406.03947","version":1},"attestation_state":"computed","paper":{"title":"Weight-based Decomposition: A Case for Bilinear MLPs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alice Rigg, Michael T. Pearce, Thomas Dooms","submitted_at":"2024-06-06T10:46:51Z","abstract_excerpt":"Gated Linear Units (GLUs) have become a common building block in modern foundation models. Bilinear layers drop the non-linearity in the \"gate\" but still have comparable performance to other GLUs. An attractive quality of bilinear layers is that they can be fully expressed in terms of a third-order tensor and linear operations. Leveraging this, we develop a method to decompose the bilinear tensor into a set of sparsely interacting eigenvectors that show promising interpretability properties in preliminary experiments for shallow image classifiers (MNIST) and small language models (Tiny Stories"},"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":"2406.03947","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-06-06T10:46:51Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"697a0a9fddc330a93fcb5fcc4ecc2764de80c3cedcb22c2e1b9b2911c083c007","abstract_canon_sha256":"3d442ba4419684e3b23770a19c28938b08df228dcdbaf429350391a728ffecc6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:28:42.465464Z","signature_b64":"RenE3ixZoEjOjv+6rC/9Cqy0iLK/CnsTW50yW3RkUzdHc3XD0lASOYk5npzueWwIFJEALDQJSJVjOXJBk0HRBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"06c0c6619e6c94fc89523825ddae9f751a85753a3a8b7a62b5406ee457644c3e","last_reissued_at":"2026-07-05T08:28:42.464971Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:28:42.464971Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Weight-based Decomposition: A Case for Bilinear MLPs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alice Rigg, Michael T. Pearce, Thomas Dooms","submitted_at":"2024-06-06T10:46:51Z","abstract_excerpt":"Gated Linear Units (GLUs) have become a common building block in modern foundation models. Bilinear layers drop the non-linearity in the \"gate\" but still have comparable performance to other GLUs. An attractive quality of bilinear layers is that they can be fully expressed in terms of a third-order tensor and linear operations. Leveraging this, we develop a method to decompose the bilinear tensor into a set of sparsely interacting eigenvectors that show promising interpretability properties in preliminary experiments for shallow image classifiers (MNIST) and small language models (Tiny Stories"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.03947","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/2406.03947/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":"2406.03947","created_at":"2026-07-05T08:28:42.465026+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.03947v1","created_at":"2026-07-05T08:28:42.465026+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.03947","created_at":"2026-07-05T08:28:42.465026+00:00"},{"alias_kind":"pith_short_12","alias_value":"A3AMMYM6NSKP","created_at":"2026-07-05T08:28:42.465026+00:00"},{"alias_kind":"pith_short_16","alias_value":"A3AMMYM6NSKPZCKS","created_at":"2026-07-05T08:28:42.465026+00:00"},{"alias_kind":"pith_short_8","alias_value":"A3AMMYM6","created_at":"2026-07-05T08:28:42.465026+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2501.16496","citing_title":"Open Problems in Mechanistic Interpretability","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/A3AMMYM6NSKPZCKSHAS53LU7OU","json":"https://pith.science/pith/A3AMMYM6NSKPZCKSHAS53LU7OU.json","graph_json":"https://pith.science/api/pith-number/A3AMMYM6NSKPZCKSHAS53LU7OU/graph.json","events_json":"https://pith.science/api/pith-number/A3AMMYM6NSKPZCKSHAS53LU7OU/events.json","paper":"https://pith.science/paper/A3AMMYM6"},"agent_actions":{"view_html":"https://pith.science/pith/A3AMMYM6NSKPZCKSHAS53LU7OU","download_json":"https://pith.science/pith/A3AMMYM6NSKPZCKSHAS53LU7OU.json","view_paper":"https://pith.science/paper/A3AMMYM6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.03947&json=true","fetch_graph":"https://pith.science/api/pith-number/A3AMMYM6NSKPZCKSHAS53LU7OU/graph.json","fetch_events":"https://pith.science/api/pith-number/A3AMMYM6NSKPZCKSHAS53LU7OU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/A3AMMYM6NSKPZCKSHAS53LU7OU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/A3AMMYM6NSKPZCKSHAS53LU7OU/action/storage_attestation","attest_author":"https://pith.science/pith/A3AMMYM6NSKPZCKSHAS53LU7OU/action/author_attestation","sign_citation":"https://pith.science/pith/A3AMMYM6NSKPZCKSHAS53LU7OU/action/citation_signature","submit_replication":"https://pith.science/pith/A3AMMYM6NSKPZCKSHAS53LU7OU/action/replication_record"}},"created_at":"2026-07-05T08:28:42.465026+00:00","updated_at":"2026-07-05T08:28:42.465026+00:00"}