{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:CCQ4XKRDFGJVD25EEVK6ATB5KZ","short_pith_number":"pith:CCQ4XKRD","schema_version":"1.0","canonical_sha256":"10a1cbaa23299351eba42555e04c3d567e64e2ee6b71a50ebb915167277a04c7","source":{"kind":"arxiv","id":"2103.03404","version":2},"attestation_state":"computed","paper":{"title":"Attention is Not All You Need: Pure Attention Loses Rank Doubly Exponentially with Depth","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Andreas Loukas, Jean-Baptiste Cordonnier, Yihe Dong","submitted_at":"2021-03-05T00:39:05Z","abstract_excerpt":"Attention-based architectures have become ubiquitous in machine learning, yet our understanding of the reasons for their effectiveness remains limited. This work proposes a new way to understand self-attention networks: we show that their output can be decomposed into a sum of smaller terms, each involving the operation of a sequence of attention heads across layers. Using this decomposition, we prove that self-attention possesses a strong inductive bias towards \"token uniformity\". Specifically, without skip connections or multi-layer perceptrons (MLPs), the output converges doubly exponential"},"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":"2103.03404","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-03-05T00:39:05Z","cross_cats_sorted":[],"title_canon_sha256":"aa558d68fe50e1494e588984af50d0bda04e8c4f5792a980360e7fb6a5ea3277","abstract_canon_sha256":"cb819f6dedd0e09e4b2608a76fa97ed0925fd6c1d9ac733ae95a396944880e73"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:36:19.015114Z","signature_b64":"j1jMhT++ZySlPZontFXzP4RJgQaKPjTr8W0CtFGZjQMg/jAajKjMyKY2gtf3jePL+YKsg/4D8yJdzddYNOjRDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"10a1cbaa23299351eba42555e04c3d567e64e2ee6b71a50ebb915167277a04c7","last_reissued_at":"2026-07-05T06:36:19.014621Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:36:19.014621Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Attention is Not All You Need: Pure Attention Loses Rank Doubly Exponentially with Depth","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Andreas Loukas, Jean-Baptiste Cordonnier, Yihe Dong","submitted_at":"2021-03-05T00:39:05Z","abstract_excerpt":"Attention-based architectures have become ubiquitous in machine learning, yet our understanding of the reasons for their effectiveness remains limited. This work proposes a new way to understand self-attention networks: we show that their output can be decomposed into a sum of smaller terms, each involving the operation of a sequence of attention heads across layers. Using this decomposition, we prove that self-attention possesses a strong inductive bias towards \"token uniformity\". Specifically, without skip connections or multi-layer perceptrons (MLPs), the output converges doubly exponential"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2103.03404","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/2103.03404/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":"2103.03404","created_at":"2026-07-05T06:36:19.014679+00:00"},{"alias_kind":"arxiv_version","alias_value":"2103.03404v2","created_at":"2026-07-05T06:36:19.014679+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2103.03404","created_at":"2026-07-05T06:36:19.014679+00:00"},{"alias_kind":"pith_short_12","alias_value":"CCQ4XKRDFGJV","created_at":"2026-07-05T06:36:19.014679+00:00"},{"alias_kind":"pith_short_16","alias_value":"CCQ4XKRDFGJVD25E","created_at":"2026-07-05T06:36:19.014679+00:00"},{"alias_kind":"pith_short_8","alias_value":"CCQ4XKRD","created_at":"2026-07-05T06:36:19.014679+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":10,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21158","citing_title":"Dead-Direction Signatures: A Cheap Spectral Reading of Singular Complexity","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2606.19491","citing_title":"Algebraic Dead Directions in LayerNorm Transformers: A Forward-Pass-Only Diagnostic at LLM Scale","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05957","citing_title":"Dead Directions: Geometric Singular Learning","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03569","citing_title":"When Attention Collapses: Stage-Aware Visual Token Pruning from Structure to Semantics","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11287","citing_title":"Beyond Similarity: Temporal Operator Attention for Time Series Analysis","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12879","citing_title":"ASAP: Amortized Doubly-Stochastic Attention via Sliced Dual Projection","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11287","citing_title":"Beyond Similarity: Temporal Operator Attention for Time Series Analysis","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09331","citing_title":"Dimension-Free Saddle-Point Escape in Muon","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07925","citing_title":"Sinkhorn doubly stochastic attention rank decay analysis","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2209.11895","citing_title":"In-context Learning and Induction Heads","ref_index":12,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CCQ4XKRDFGJVD25EEVK6ATB5KZ","json":"https://pith.science/pith/CCQ4XKRDFGJVD25EEVK6ATB5KZ.json","graph_json":"https://pith.science/api/pith-number/CCQ4XKRDFGJVD25EEVK6ATB5KZ/graph.json","events_json":"https://pith.science/api/pith-number/CCQ4XKRDFGJVD25EEVK6ATB5KZ/events.json","paper":"https://pith.science/paper/CCQ4XKRD"},"agent_actions":{"view_html":"https://pith.science/pith/CCQ4XKRDFGJVD25EEVK6ATB5KZ","download_json":"https://pith.science/pith/CCQ4XKRDFGJVD25EEVK6ATB5KZ.json","view_paper":"https://pith.science/paper/CCQ4XKRD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2103.03404&json=true","fetch_graph":"https://pith.science/api/pith-number/CCQ4XKRDFGJVD25EEVK6ATB5KZ/graph.json","fetch_events":"https://pith.science/api/pith-number/CCQ4XKRDFGJVD25EEVK6ATB5KZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CCQ4XKRDFGJVD25EEVK6ATB5KZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CCQ4XKRDFGJVD25EEVK6ATB5KZ/action/storage_attestation","attest_author":"https://pith.science/pith/CCQ4XKRDFGJVD25EEVK6ATB5KZ/action/author_attestation","sign_citation":"https://pith.science/pith/CCQ4XKRDFGJVD25EEVK6ATB5KZ/action/citation_signature","submit_replication":"https://pith.science/pith/CCQ4XKRDFGJVD25EEVK6ATB5KZ/action/replication_record"}},"created_at":"2026-07-05T06:36:19.014679+00:00","updated_at":"2026-07-05T06:36:19.014679+00:00"}