{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:6Y2K6T2GG7UNX2WPNIURXEHKME","short_pith_number":"pith:6Y2K6T2G","schema_version":"1.0","canonical_sha256":"f634af4f4637e8dbeacf6a291b90ea6115b50f43012b49cefdb8bbe39197592f","source":{"kind":"arxiv","id":"2605.20271","version":1},"attestation_state":"computed","paper":{"title":"Multi-Head Attention as Ensemble Nadaraya-Watson Estimation: Variance Reduction, Decorrelation, and Optimal Head Diversity","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Ernest Fokou\\'e","submitted_at":"2026-05-18T23:43:53Z","abstract_excerpt":"We develop a rigorous statistical theory of multi-head attention (MHA) as an ensemble of Nadaraya-Watson (NW) kernel regression estimators. Building on the algebraic identity between single-head softmax attention and the NW estimator, we prove that MHA is a structured ensemble of H NW estimators, each operating in a distinct learned projection subspace of the key space. We derive an explicit Bias-Variance-Covariance decomposition of the MHA mean squared error, showing that variance reduction depends not merely on the number of heads H but fundamentally on the decorrelation of head outputs. Dec"},"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":"2605.20271","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2026-05-18T23:43:53Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"2646223115b2d96ac7aa89aed52ad45b7d40f353c8a725aca2a896de7abdae79","abstract_canon_sha256":"1433f670b5f9dd997bd6074186dce61d240e4bae8dcad4911a8cd19c4bbb2a30"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-21T00:04:23.493325Z","signature_b64":"bW2lOL3L9rFf9Abc5NmtRMfue/MWywrQEvrXnWofZXfGq+O52ssoqcDrDOXbB47WTCCT+ZNBQg2gB8j4nYr3DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f634af4f4637e8dbeacf6a291b90ea6115b50f43012b49cefdb8bbe39197592f","last_reissued_at":"2026-05-21T00:04:23.492719Z","signature_status":"signed_v1","first_computed_at":"2026-05-21T00:04:23.492719Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Multi-Head Attention as Ensemble Nadaraya-Watson Estimation: Variance Reduction, Decorrelation, and Optimal Head Diversity","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Ernest Fokou\\'e","submitted_at":"2026-05-18T23:43:53Z","abstract_excerpt":"We develop a rigorous statistical theory of multi-head attention (MHA) as an ensemble of Nadaraya-Watson (NW) kernel regression estimators. Building on the algebraic identity between single-head softmax attention and the NW estimator, we prove that MHA is a structured ensemble of H NW estimators, each operating in a distinct learned projection subspace of the key space. We derive an explicit Bias-Variance-Covariance decomposition of the MHA mean squared error, showing that variance reduction depends not merely on the number of heads H but fundamentally on the decorrelation of head outputs. Dec"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.20271","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/2605.20271/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":"2605.20271","created_at":"2026-05-21T00:04:23.492788+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.20271v1","created_at":"2026-05-21T00:04:23.492788+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.20271","created_at":"2026-05-21T00:04:23.492788+00:00"},{"alias_kind":"pith_short_12","alias_value":"6Y2K6T2GG7UN","created_at":"2026-05-21T00:04:23.492788+00:00"},{"alias_kind":"pith_short_16","alias_value":"6Y2K6T2GG7UNX2WP","created_at":"2026-05-21T00:04:23.492788+00:00"},{"alias_kind":"pith_short_8","alias_value":"6Y2K6T2G","created_at":"2026-05-21T00:04:23.492788+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/6Y2K6T2GG7UNX2WPNIURXEHKME","json":"https://pith.science/pith/6Y2K6T2GG7UNX2WPNIURXEHKME.json","graph_json":"https://pith.science/api/pith-number/6Y2K6T2GG7UNX2WPNIURXEHKME/graph.json","events_json":"https://pith.science/api/pith-number/6Y2K6T2GG7UNX2WPNIURXEHKME/events.json","paper":"https://pith.science/paper/6Y2K6T2G"},"agent_actions":{"view_html":"https://pith.science/pith/6Y2K6T2GG7UNX2WPNIURXEHKME","download_json":"https://pith.science/pith/6Y2K6T2GG7UNX2WPNIURXEHKME.json","view_paper":"https://pith.science/paper/6Y2K6T2G","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.20271&json=true","fetch_graph":"https://pith.science/api/pith-number/6Y2K6T2GG7UNX2WPNIURXEHKME/graph.json","fetch_events":"https://pith.science/api/pith-number/6Y2K6T2GG7UNX2WPNIURXEHKME/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6Y2K6T2GG7UNX2WPNIURXEHKME/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6Y2K6T2GG7UNX2WPNIURXEHKME/action/storage_attestation","attest_author":"https://pith.science/pith/6Y2K6T2GG7UNX2WPNIURXEHKME/action/author_attestation","sign_citation":"https://pith.science/pith/6Y2K6T2GG7UNX2WPNIURXEHKME/action/citation_signature","submit_replication":"https://pith.science/pith/6Y2K6T2GG7UNX2WPNIURXEHKME/action/replication_record"}},"created_at":"2026-05-21T00:04:23.492788+00:00","updated_at":"2026-05-21T00:04:23.492788+00:00"}