{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:FTN3NEOMYNL2BWVRKLITK4ZZBN","short_pith_number":"pith:FTN3NEOM","schema_version":"1.0","canonical_sha256":"2cdbb691ccc357a0dab152d13573390b7b1011190a79132b3a34d73aa31e707b","source":{"kind":"arxiv","id":"2104.07012","version":2},"attestation_state":"computed","paper":{"title":"Sparse Attention with Linear Units","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Biao Zhang, Ivan Titov, Rico Sennrich","submitted_at":"2021-04-14T17:52:38Z","abstract_excerpt":"Recently, it has been argued that encoder-decoder models can be made more interpretable by replacing the softmax function in the attention with its sparse variants. In this work, we introduce a novel, simple method for achieving sparsity in attention: we replace the softmax activation with a ReLU, and show that sparsity naturally emerges from such a formulation. Training stability is achieved with layer normalization with either a specialized initialization or an additional gating function. Our model, which we call Rectified Linear Attention (ReLA), is easy to implement and more efficient than"},"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":"2104.07012","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2021-04-14T17:52:38Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"5cbb3cd2894e8c360c3d48eb1922b65aedf8f4663aae898a5f1304791bea66fa","abstract_canon_sha256":"28a6aa98abb456d522ad3b86ace6647a50a6e10d3189d4deaf123b67ea724381"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:20:20.559163Z","signature_b64":"yxLgIfD3kqgsnTQBoke5ZJDsJirny8j4PAxlABwJhkPRGIK+ZWPxDm7dXaIw0TltME4agToucrcg3hkE/ercAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2cdbb691ccc357a0dab152d13573390b7b1011190a79132b3a34d73aa31e707b","last_reissued_at":"2026-07-05T03:20:20.558767Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:20:20.558767Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Sparse Attention with Linear Units","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Biao Zhang, Ivan Titov, Rico Sennrich","submitted_at":"2021-04-14T17:52:38Z","abstract_excerpt":"Recently, it has been argued that encoder-decoder models can be made more interpretable by replacing the softmax function in the attention with its sparse variants. In this work, we introduce a novel, simple method for achieving sparsity in attention: we replace the softmax activation with a ReLU, and show that sparsity naturally emerges from such a formulation. Training stability is achieved with layer normalization with either a specialized initialization or an additional gating function. Our model, which we call Rectified Linear Attention (ReLA), is easy to implement and more efficient than"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2104.07012","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/2104.07012/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":"2104.07012","created_at":"2026-07-05T03:20:20.558818+00:00"},{"alias_kind":"arxiv_version","alias_value":"2104.07012v2","created_at":"2026-07-05T03:20:20.558818+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2104.07012","created_at":"2026-07-05T03:20:20.558818+00:00"},{"alias_kind":"pith_short_12","alias_value":"FTN3NEOMYNL2","created_at":"2026-07-05T03:20:20.558818+00:00"},{"alias_kind":"pith_short_16","alias_value":"FTN3NEOMYNL2BWVR","created_at":"2026-07-05T03:20:20.558818+00:00"},{"alias_kind":"pith_short_8","alias_value":"FTN3NEOM","created_at":"2026-07-05T03:20:20.558818+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.31609","citing_title":"Learning Structurally Consistent Representations for Multi-View Radar Semantic Segmentation","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10198","citing_title":"Empty SPACE: Cross-Attention Sparsity for Concept Erasure in Diffusion Models","ref_index":38,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FTN3NEOMYNL2BWVRKLITK4ZZBN","json":"https://pith.science/pith/FTN3NEOMYNL2BWVRKLITK4ZZBN.json","graph_json":"https://pith.science/api/pith-number/FTN3NEOMYNL2BWVRKLITK4ZZBN/graph.json","events_json":"https://pith.science/api/pith-number/FTN3NEOMYNL2BWVRKLITK4ZZBN/events.json","paper":"https://pith.science/paper/FTN3NEOM"},"agent_actions":{"view_html":"https://pith.science/pith/FTN3NEOMYNL2BWVRKLITK4ZZBN","download_json":"https://pith.science/pith/FTN3NEOMYNL2BWVRKLITK4ZZBN.json","view_paper":"https://pith.science/paper/FTN3NEOM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2104.07012&json=true","fetch_graph":"https://pith.science/api/pith-number/FTN3NEOMYNL2BWVRKLITK4ZZBN/graph.json","fetch_events":"https://pith.science/api/pith-number/FTN3NEOMYNL2BWVRKLITK4ZZBN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FTN3NEOMYNL2BWVRKLITK4ZZBN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FTN3NEOMYNL2BWVRKLITK4ZZBN/action/storage_attestation","attest_author":"https://pith.science/pith/FTN3NEOMYNL2BWVRKLITK4ZZBN/action/author_attestation","sign_citation":"https://pith.science/pith/FTN3NEOMYNL2BWVRKLITK4ZZBN/action/citation_signature","submit_replication":"https://pith.science/pith/FTN3NEOMYNL2BWVRKLITK4ZZBN/action/replication_record"}},"created_at":"2026-07-05T03:20:20.558818+00:00","updated_at":"2026-07-05T03:20:20.558818+00:00"}