{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:LFH4ZRFLEHTI75TND2PJ4Q4OTL","short_pith_number":"pith:LFH4ZRFL","schema_version":"1.0","canonical_sha256":"594fccc4ab21e68ff66d1e9e9e438e9ae7ed727da50b5108771551f1c8da0fbb","source":{"kind":"arxiv","id":"2209.10655","version":3},"attestation_state":"computed","paper":{"title":"Mega: Moving Average Equipped Gated Attention","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Chunting Zhou, Graham Neubig, Jonathan May, Junxian He, Liangke Gui, Luke Zettlemoyer, Xiang Kong, Xuezhe Ma","submitted_at":"2022-09-21T20:52:17Z","abstract_excerpt":"The design choices in the Transformer attention mechanism, including weak inductive bias and quadratic computational complexity, have limited its application for modeling long sequences. In this paper, we introduce Mega, a simple, theoretically grounded, single-head gated attention mechanism equipped with (exponential) moving average to incorporate inductive bias of position-aware local dependencies into the position-agnostic attention mechanism. We further propose a variant of Mega that offers linear time and space complexity yet yields only minimal quality loss, by efficiently splitting the "},"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":"2209.10655","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2022-09-21T20:52:17Z","cross_cats_sorted":[],"title_canon_sha256":"e9a385c2acd7a7857c098b33274b84c7b4b2a83de554360f683e0cacbb4c7604","abstract_canon_sha256":"4ca5b3247e5ea330f4906426b11a9e8620110a1a1af0405dad72490d165beddb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:36:31.626469Z","signature_b64":"T2trcgMHETTCf4sWbyRZeyhEhLSrbsXLDurlN1YOL7uSkr6Cap5LQUFyeIj3168ZT3VpFUBpK8MR7dhN+Y3xCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"594fccc4ab21e68ff66d1e9e9e438e9ae7ed727da50b5108771551f1c8da0fbb","last_reissued_at":"2026-07-05T05:36:31.625910Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:36:31.625910Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mega: Moving Average Equipped Gated Attention","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Chunting Zhou, Graham Neubig, Jonathan May, Junxian He, Liangke Gui, Luke Zettlemoyer, Xiang Kong, Xuezhe Ma","submitted_at":"2022-09-21T20:52:17Z","abstract_excerpt":"The design choices in the Transformer attention mechanism, including weak inductive bias and quadratic computational complexity, have limited its application for modeling long sequences. In this paper, we introduce Mega, a simple, theoretically grounded, single-head gated attention mechanism equipped with (exponential) moving average to incorporate inductive bias of position-aware local dependencies into the position-agnostic attention mechanism. We further propose a variant of Mega that offers linear time and space complexity yet yields only minimal quality loss, by efficiently splitting the "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2209.10655","kind":"arxiv","version":3},"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/2209.10655/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":"2209.10655","created_at":"2026-07-05T05:36:31.625971+00:00"},{"alias_kind":"arxiv_version","alias_value":"2209.10655v3","created_at":"2026-07-05T05:36:31.625971+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2209.10655","created_at":"2026-07-05T05:36:31.625971+00:00"},{"alias_kind":"pith_short_12","alias_value":"LFH4ZRFLEHTI","created_at":"2026-07-05T05:36:31.625971+00:00"},{"alias_kind":"pith_short_16","alias_value":"LFH4ZRFLEHTI75TN","created_at":"2026-07-05T05:36:31.625971+00:00"},{"alias_kind":"pith_short_8","alias_value":"LFH4ZRFL","created_at":"2026-07-05T05:36:31.625971+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.10802","citing_title":"Boosting ECG Classification Performance by Pre-training with Synthesized Data","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2312.02549","citing_title":"DemaFormer: Damped Exponential Moving Average Transformer with Energy-Based Modeling for Temporal Language Grounding","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2503.18970","citing_title":"Advancing Intelligent Sequence Modeling: Evolution, Trade-offs, and Applications of State- Space Architectures from S4 to Mamba","ref_index":73,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21070","citing_title":"Towards Understanding Self-Pretraining for Sequence Classification","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2511.08877","citing_title":"Hierarchical Memorization in Large Language Models: Evidence from Citation Generation","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2512.12602","citing_title":"Exact Flow Linear Attention: Exact Solution from Continuous-Time Dynamics","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05066","citing_title":"The Impossibility Triangle of Long-Context Modeling","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LFH4ZRFLEHTI75TND2PJ4Q4OTL","json":"https://pith.science/pith/LFH4ZRFLEHTI75TND2PJ4Q4OTL.json","graph_json":"https://pith.science/api/pith-number/LFH4ZRFLEHTI75TND2PJ4Q4OTL/graph.json","events_json":"https://pith.science/api/pith-number/LFH4ZRFLEHTI75TND2PJ4Q4OTL/events.json","paper":"https://pith.science/paper/LFH4ZRFL"},"agent_actions":{"view_html":"https://pith.science/pith/LFH4ZRFLEHTI75TND2PJ4Q4OTL","download_json":"https://pith.science/pith/LFH4ZRFLEHTI75TND2PJ4Q4OTL.json","view_paper":"https://pith.science/paper/LFH4ZRFL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2209.10655&json=true","fetch_graph":"https://pith.science/api/pith-number/LFH4ZRFLEHTI75TND2PJ4Q4OTL/graph.json","fetch_events":"https://pith.science/api/pith-number/LFH4ZRFLEHTI75TND2PJ4Q4OTL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LFH4ZRFLEHTI75TND2PJ4Q4OTL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LFH4ZRFLEHTI75TND2PJ4Q4OTL/action/storage_attestation","attest_author":"https://pith.science/pith/LFH4ZRFLEHTI75TND2PJ4Q4OTL/action/author_attestation","sign_citation":"https://pith.science/pith/LFH4ZRFLEHTI75TND2PJ4Q4OTL/action/citation_signature","submit_replication":"https://pith.science/pith/LFH4ZRFLEHTI75TND2PJ4Q4OTL/action/replication_record"}},"created_at":"2026-07-05T05:36:31.625971+00:00","updated_at":"2026-07-05T05:36:31.625971+00:00"}