{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:YUHHGITECURLB5LIIQHHCYNJJX","short_pith_number":"pith:YUHHGITE","schema_version":"1.0","canonical_sha256":"c50e7322641522b0f568440e7161a94dd7f870ccf784b749c79e49c33e6dcdea","source":{"kind":"arxiv","id":"2405.08553","version":2},"attestation_state":"computed","paper":{"title":"Improving Transformers with Dynamically Composable Multi-Head Attention","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Da Xiao, Qingye Meng, Shengping Li, Xingyuan Yuan","submitted_at":"2024-05-14T12:41:11Z","abstract_excerpt":"Multi-Head Attention (MHA) is a key component of Transformer. In MHA, attention heads work independently, causing problems such as low-rank bottleneck of attention score matrices and head redundancy. We propose Dynamically Composable Multi-Head Attention (DCMHA), a parameter and computation efficient attention architecture that tackles the shortcomings of MHA and increases the expressive power of the model by dynamically composing attention heads. At the core of DCMHA is a $\\it{Compose}$ function that transforms the attention score and weight matrices in an input-dependent way. DCMHA can be us"},"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":"2405.08553","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-05-14T12:41:11Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"171e0daefd5860d4aaf4ef70a7d3c21b2505cecbb762b8d210a0f46d28fffa93","abstract_canon_sha256":"56aa167f93cc776307d2d75c8cf2914c361d7c84f46121742df2c1280c0f4f01"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:27:08.187705Z","signature_b64":"ANfPlRqE6hiqTPkWWRKy94wk8kAEQMwmeZs9aWhpyNgErC2Ug9q+yE2QxmJmIqCBa1B0CdLQEcguzA682MiZAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c50e7322641522b0f568440e7161a94dd7f870ccf784b749c79e49c33e6dcdea","last_reissued_at":"2026-07-05T08:27:08.187245Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:27:08.187245Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Improving Transformers with Dynamically Composable Multi-Head Attention","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Da Xiao, Qingye Meng, Shengping Li, Xingyuan Yuan","submitted_at":"2024-05-14T12:41:11Z","abstract_excerpt":"Multi-Head Attention (MHA) is a key component of Transformer. In MHA, attention heads work independently, causing problems such as low-rank bottleneck of attention score matrices and head redundancy. We propose Dynamically Composable Multi-Head Attention (DCMHA), a parameter and computation efficient attention architecture that tackles the shortcomings of MHA and increases the expressive power of the model by dynamically composing attention heads. At the core of DCMHA is a $\\it{Compose}$ function that transforms the attention score and weight matrices in an input-dependent way. DCMHA can be us"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.08553","kind":"arxiv","version":2},"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/2405.08553/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":"2405.08553","created_at":"2026-07-05T08:27:08.187300+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.08553v2","created_at":"2026-07-05T08:27:08.187300+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.08553","created_at":"2026-07-05T08:27:08.187300+00:00"},{"alias_kind":"pith_short_12","alias_value":"YUHHGITECURL","created_at":"2026-07-05T08:27:08.187300+00:00"},{"alias_kind":"pith_short_16","alias_value":"YUHHGITECURLB5LI","created_at":"2026-07-05T08:27:08.187300+00:00"},{"alias_kind":"pith_short_8","alias_value":"YUHHGITE","created_at":"2026-07-05T08:27:08.187300+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.07604","citing_title":"Contribution Weights: A Geometrical Analysis of Self-Attention Transformers","ref_index":135,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YUHHGITECURLB5LIIQHHCYNJJX","json":"https://pith.science/pith/YUHHGITECURLB5LIIQHHCYNJJX.json","graph_json":"https://pith.science/api/pith-number/YUHHGITECURLB5LIIQHHCYNJJX/graph.json","events_json":"https://pith.science/api/pith-number/YUHHGITECURLB5LIIQHHCYNJJX/events.json","paper":"https://pith.science/paper/YUHHGITE"},"agent_actions":{"view_html":"https://pith.science/pith/YUHHGITECURLB5LIIQHHCYNJJX","download_json":"https://pith.science/pith/YUHHGITECURLB5LIIQHHCYNJJX.json","view_paper":"https://pith.science/paper/YUHHGITE","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.08553&json=true","fetch_graph":"https://pith.science/api/pith-number/YUHHGITECURLB5LIIQHHCYNJJX/graph.json","fetch_events":"https://pith.science/api/pith-number/YUHHGITECURLB5LIIQHHCYNJJX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YUHHGITECURLB5LIIQHHCYNJJX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YUHHGITECURLB5LIIQHHCYNJJX/action/storage_attestation","attest_author":"https://pith.science/pith/YUHHGITECURLB5LIIQHHCYNJJX/action/author_attestation","sign_citation":"https://pith.science/pith/YUHHGITECURLB5LIIQHHCYNJJX/action/citation_signature","submit_replication":"https://pith.science/pith/YUHHGITECURLB5LIIQHHCYNJJX/action/replication_record"}},"created_at":"2026-07-05T08:27:08.187300+00:00","updated_at":"2026-07-05T08:27:08.187300+00:00"}