{"paper":{"title":"Learning Subspace-Preserving Sparse Attention Graphs from Heterogeneous Multiview Data","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A sparse attention graph learning method recovers subspace structures from heterogeneous multiview data using bilinear factorization and entmax projections.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Chuanbin Liu, Jie Chen, Xi Peng, Yuanbiao Gou, Zhu Wang","submitted_at":"2026-05-12T09:56:28Z","abstract_excerpt":"The high-dimensional features extracted from large-scale unlabeled data via various pretrained models with diverse architectures are referred to as heterogeneous multiview data. Most existing unsupervised transfer learning methods fail to faithfully recover intrinsic subspace structures when exploiting complementary information across multiple views. Therefore, a fundamental challenge involves constructing sparse similarity graphs that preserve these underlying subspace structures for achieving semantic alignment across heterogeneous views. In this paper, we propose a sparse attention graph le"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"we propose a sparse attention graph learning (SAGL) method that learns subspace-preserving sparse attention graphs from heterogeneous multiview data... SAGL consistently outperforms the state-of-the-art unsupervised transfer learning approaches.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the bilinear attention factorization combined with α-entmax projection and dynamic sparsity gating will faithfully recover intrinsic subspace structures across heterogeneous views without introducing artifacts that harm semantic alignment.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SAGL learns subspace-preserving sparse attention graphs from heterogeneous multiview data via bilinear attention factorization, dynamic sparsity gating, and α-entmax projection.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A sparse attention graph learning method recovers subspace structures from heterogeneous multiview data using bilinear factorization and entmax projections.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"9b379bd22a512795004d4edbe2f86d116a9b57f3d4565aebc4508585243e788c"},"source":{"id":"2605.11881","kind":"arxiv","version":2},"verdict":{"id":"3708c085-8991-42ca-a19f-0066f8bd0030","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T06:58:03.848877Z","strongest_claim":"we propose a sparse attention graph learning (SAGL) method that learns subspace-preserving sparse attention graphs from heterogeneous multiview data... SAGL consistently outperforms the state-of-the-art unsupervised transfer learning approaches.","one_line_summary":"SAGL learns subspace-preserving sparse attention graphs from heterogeneous multiview data via bilinear attention factorization, dynamic sparsity gating, and α-entmax projection.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the bilinear attention factorization combined with α-entmax projection and dynamic sparsity gating will faithfully recover intrinsic subspace structures across heterogeneous views without introducing artifacts that harm semantic alignment.","pith_extraction_headline":"A sparse attention graph learning method recovers subspace structures from heterogeneous multiview data using bilinear factorization and entmax projections."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.11881/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-19T11:36:24.410748Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T09:01:18.023437Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T08:05:39.044352Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"a0c1d4047a2083e34ebba4ae8f3608c67c6f6e114415ee0a43d9bde66b71a68c"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"1df1683b968b73f6da8151cbadc8cd964ad1048023c4d137258fcbd5665c8e0d"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}