{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:6ZXVA4T5GSLU54KIVRZ2LQ4SYA","short_pith_number":"pith:6ZXVA4T5","schema_version":"1.0","canonical_sha256":"f66f50727d34974ef148ac73a5c392c02115fbc98ee1445843a02c5728beebde","source":{"kind":"arxiv","id":"2408.09624","version":1},"attestation_state":"computed","paper":{"title":"Attention is a smoothed cubic spline","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG","cs.NA","math.NA"],"primary_cat":"cs.AI","authors_text":"Lek-Heng Lim, Yucong Liu, Zehua Lai","submitted_at":"2024-08-19T00:56:44Z","abstract_excerpt":"We highlight a perhaps important but hitherto unobserved insight: The attention module in a transformer is a smoothed cubic spline. Viewed in this manner, this mysterious but critical component of a transformer becomes a natural development of an old notion deeply entrenched in classical approximation theory. More precisely, we show that with ReLU-activation, attention, masked attention, encoder-decoder attention are all cubic splines. As every component in a transformer is constructed out of compositions of various attention modules (= cubic splines) and feed forward neural networks (= linear"},"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":"2408.09624","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2024-08-19T00:56:44Z","cross_cats_sorted":["cs.LG","cs.NA","math.NA"],"title_canon_sha256":"f48c6486bf13bb933b5c514d939aabc89653b9a741aae9b988e942d5b23bb7c0","abstract_canon_sha256":"f4295901b02d95f529566f4a713ed8c86468e7578a19c08366c092241fa1408f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:56:34.778275Z","signature_b64":"8rFTmCA4D0+Op2dbXx9ZsxD3q474lMISrK4jhC/x3b2sxPR36dnTovgpsBtBNrhu1pnQa3gvKGL49PbjbEwrBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f66f50727d34974ef148ac73a5c392c02115fbc98ee1445843a02c5728beebde","last_reissued_at":"2026-07-05T08:56:34.777802Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:56:34.777802Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Attention is a smoothed cubic spline","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG","cs.NA","math.NA"],"primary_cat":"cs.AI","authors_text":"Lek-Heng Lim, Yucong Liu, Zehua Lai","submitted_at":"2024-08-19T00:56:44Z","abstract_excerpt":"We highlight a perhaps important but hitherto unobserved insight: The attention module in a transformer is a smoothed cubic spline. Viewed in this manner, this mysterious but critical component of a transformer becomes a natural development of an old notion deeply entrenched in classical approximation theory. More precisely, we show that with ReLU-activation, attention, masked attention, encoder-decoder attention are all cubic splines. As every component in a transformer is constructed out of compositions of various attention modules (= cubic splines) and feed forward neural networks (= linear"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.09624","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/2408.09624/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":"2408.09624","created_at":"2026-07-05T08:56:34.777863+00:00"},{"alias_kind":"arxiv_version","alias_value":"2408.09624v1","created_at":"2026-07-05T08:56:34.777863+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.09624","created_at":"2026-07-05T08:56:34.777863+00:00"},{"alias_kind":"pith_short_12","alias_value":"6ZXVA4T5GSLU","created_at":"2026-07-05T08:56:34.777863+00:00"},{"alias_kind":"pith_short_16","alias_value":"6ZXVA4T5GSLU54KI","created_at":"2026-07-05T08:56:34.777863+00:00"},{"alias_kind":"pith_short_8","alias_value":"6ZXVA4T5","created_at":"2026-07-05T08:56:34.777863+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2505.03205","citing_title":"Transformers for Learning on Noisy and Task-Level Manifolds: Approximation and Generalization Insights","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15970","citing_title":"Copositive Matrices with Ordered Off-Diagonal Entries","ref_index":208,"is_internal_anchor":false},{"citing_arxiv_id":"2510.03989","citing_title":"A Mathematical Explanation of Transformers","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15632","citing_title":"Algebraic Invariants of Lightning Self-Attention","ref_index":26,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6ZXVA4T5GSLU54KIVRZ2LQ4SYA","json":"https://pith.science/pith/6ZXVA4T5GSLU54KIVRZ2LQ4SYA.json","graph_json":"https://pith.science/api/pith-number/6ZXVA4T5GSLU54KIVRZ2LQ4SYA/graph.json","events_json":"https://pith.science/api/pith-number/6ZXVA4T5GSLU54KIVRZ2LQ4SYA/events.json","paper":"https://pith.science/paper/6ZXVA4T5"},"agent_actions":{"view_html":"https://pith.science/pith/6ZXVA4T5GSLU54KIVRZ2LQ4SYA","download_json":"https://pith.science/pith/6ZXVA4T5GSLU54KIVRZ2LQ4SYA.json","view_paper":"https://pith.science/paper/6ZXVA4T5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2408.09624&json=true","fetch_graph":"https://pith.science/api/pith-number/6ZXVA4T5GSLU54KIVRZ2LQ4SYA/graph.json","fetch_events":"https://pith.science/api/pith-number/6ZXVA4T5GSLU54KIVRZ2LQ4SYA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6ZXVA4T5GSLU54KIVRZ2LQ4SYA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6ZXVA4T5GSLU54KIVRZ2LQ4SYA/action/storage_attestation","attest_author":"https://pith.science/pith/6ZXVA4T5GSLU54KIVRZ2LQ4SYA/action/author_attestation","sign_citation":"https://pith.science/pith/6ZXVA4T5GSLU54KIVRZ2LQ4SYA/action/citation_signature","submit_replication":"https://pith.science/pith/6ZXVA4T5GSLU54KIVRZ2LQ4SYA/action/replication_record"}},"created_at":"2026-07-05T08:56:34.777863+00:00","updated_at":"2026-07-05T08:56:34.777863+00:00"}