{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:HAA2MNCFOEPKJ6WACTCEHJS6FF","short_pith_number":"pith:HAA2MNCF","schema_version":"1.0","canonical_sha256":"3801a63445711ea4fac014c443a65e294c64caca53c5f50e37116fd914cb67b4","source":{"kind":"arxiv","id":"2608.07616","version":1},"attestation_state":"computed","paper":{"title":"HSMLA: Hierarchical Softmax Multi-scale Linear Attention for Efficient Vision Transformers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Dong Liu, Renata Borovica-Gajic, Yanxuan Yu, Ying Nian Wu","submitted_at":"2026-08-07T03:47:04Z","abstract_excerpt":"Vision transformers face significant computational overheads in high-resolution dense prediction due to the quadratic complexity of self-attention. Linear attention offers efficiency but sacrifices local context modeling. We propose \\textbf{HSMLA (Hierarchical Softmax Multi-scale Linear Attention)}, which combines ReLU-based linear attention for global context, selective softmax refinement for critical local features, and multi-scale token representations via depthwise convolutions. HSMLA achieves superior accuracy-efficiency trade-offs: up to $4.2\\times$ inference-time speedup across dense pr"},"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":"2608.07616","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-08-07T03:47:04Z","cross_cats_sorted":[],"title_canon_sha256":"7e256c2afee13391fffe4e22bb6d7df93264850c3bb55da2d2945438372b2aab","abstract_canon_sha256":"fa4fb5cc4dc7f9c79210e456955efd19242dd781f87e6ae3fb5529dbc395f2f4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-11T00:15:29.890946Z","signature_b64":"5VA9RuMJZaOSrZYtD8ZMWy9rpM1qHEERmxJvtbph5ug+x8/ZNrrFEIH/4xkG4cjIhRUpvAZo3buUkHB6KTPqCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3801a63445711ea4fac014c443a65e294c64caca53c5f50e37116fd914cb67b4","last_reissued_at":"2026-08-11T00:15:29.888774Z","signature_status":"signed_v1","first_computed_at":"2026-08-11T00:15:29.888774Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"HSMLA: Hierarchical Softmax Multi-scale Linear Attention for Efficient Vision Transformers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Dong Liu, Renata Borovica-Gajic, Yanxuan Yu, Ying Nian Wu","submitted_at":"2026-08-07T03:47:04Z","abstract_excerpt":"Vision transformers face significant computational overheads in high-resolution dense prediction due to the quadratic complexity of self-attention. Linear attention offers efficiency but sacrifices local context modeling. We propose \\textbf{HSMLA (Hierarchical Softmax Multi-scale Linear Attention)}, which combines ReLU-based linear attention for global context, selective softmax refinement for critical local features, and multi-scale token representations via depthwise convolutions. HSMLA achieves superior accuracy-efficiency trade-offs: up to $4.2\\times$ inference-time speedup across dense pr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.07616","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/2608.07616/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":"2608.07616","created_at":"2026-08-11T00:15:29.889125+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.07616v1","created_at":"2026-08-11T00:15:29.889125+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.07616","created_at":"2026-08-11T00:15:29.889125+00:00"},{"alias_kind":"pith_short_12","alias_value":"HAA2MNCFOEPK","created_at":"2026-08-11T00:15:29.889125+00:00"},{"alias_kind":"pith_short_16","alias_value":"HAA2MNCFOEPKJ6WA","created_at":"2026-08-11T00:15:29.889125+00:00"},{"alias_kind":"pith_short_8","alias_value":"HAA2MNCF","created_at":"2026-08-11T00:15:29.889125+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/HAA2MNCFOEPKJ6WACTCEHJS6FF","json":"https://pith.science/pith/HAA2MNCFOEPKJ6WACTCEHJS6FF.json","graph_json":"https://pith.science/api/pith-number/HAA2MNCFOEPKJ6WACTCEHJS6FF/graph.json","events_json":"https://pith.science/api/pith-number/HAA2MNCFOEPKJ6WACTCEHJS6FF/events.json","paper":"https://pith.science/paper/HAA2MNCF"},"agent_actions":{"view_html":"https://pith.science/pith/HAA2MNCFOEPKJ6WACTCEHJS6FF","download_json":"https://pith.science/pith/HAA2MNCFOEPKJ6WACTCEHJS6FF.json","view_paper":"https://pith.science/paper/HAA2MNCF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.07616&json=true","fetch_graph":"https://pith.science/api/pith-number/HAA2MNCFOEPKJ6WACTCEHJS6FF/graph.json","fetch_events":"https://pith.science/api/pith-number/HAA2MNCFOEPKJ6WACTCEHJS6FF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HAA2MNCFOEPKJ6WACTCEHJS6FF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HAA2MNCFOEPKJ6WACTCEHJS6FF/action/storage_attestation","attest_author":"https://pith.science/pith/HAA2MNCFOEPKJ6WACTCEHJS6FF/action/author_attestation","sign_citation":"https://pith.science/pith/HAA2MNCFOEPKJ6WACTCEHJS6FF/action/citation_signature","submit_replication":"https://pith.science/pith/HAA2MNCFOEPKJ6WACTCEHJS6FF/action/replication_record"}},"created_at":"2026-08-11T00:15:29.889125+00:00","updated_at":"2026-08-11T00:15:29.889125+00:00"}