{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:G2UAMCOBO7R7B5PI2FJJEZQNMZ","short_pith_number":"pith:G2UAMCOB","schema_version":"1.0","canonical_sha256":"36a80609c177e3f0f5e8d15292660d66416ce7b6dd17a4d9410f8f346853bac9","source":{"kind":"arxiv","id":"2006.12467","version":3},"attestation_state":"computed","paper":{"title":"The Depth-to-Width Interplay in Self-Attention","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"","cross_cats":["cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Amnon Shashua, Hofit Bata, Noam Wies, Or Sharir, Yoav Levine","submitted_at":"2020-06-22T17:47:09Z","abstract_excerpt":"Self-attention architectures, which are rapidly pushing the frontier in natural language processing, demonstrate a surprising depth-inefficient behavior: previous works indicate that increasing the internal representation (network width) is just as useful as increasing the number of self-attention layers (network depth). We theoretically predict a width-dependent transition between depth-efficiency and depth-inefficiency in self-attention. We conduct systematic empirical ablations on networks of depths 6 to 48 that clearly reveal the theoretically predicted behaviors, and provide explicit quan"},"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":"2006.12467","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/publicdomain/zero/1.0/","primary_cat":"cs.LG","submitted_at":"2020-06-22T17:47:09Z","cross_cats_sorted":["cs.CL","stat.ML"],"title_canon_sha256":"9f7bfc8414c44a1d45db724f55b954d559c1bb8e8fa77c13f5c823e0edbaacdf","abstract_canon_sha256":"5778acc552de6c2206d574aeffd20f33887c3f35eb1826b6649f88ca3035d2e5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:07:28.909819Z","signature_b64":"95bDaCdhNIpa5rG1pUgaXZrIkCc8mvpcj67HzTN/TycgvT0ewWFj2Q9DpB+V16Nm3KotiiT6hsLL/MALCKAhAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"36a80609c177e3f0f5e8d15292660d66416ce7b6dd17a4d9410f8f346853bac9","last_reissued_at":"2026-07-05T02:07:28.909291Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:07:28.909291Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Depth-to-Width Interplay in Self-Attention","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"","cross_cats":["cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Amnon Shashua, Hofit Bata, Noam Wies, Or Sharir, Yoav Levine","submitted_at":"2020-06-22T17:47:09Z","abstract_excerpt":"Self-attention architectures, which are rapidly pushing the frontier in natural language processing, demonstrate a surprising depth-inefficient behavior: previous works indicate that increasing the internal representation (network width) is just as useful as increasing the number of self-attention layers (network depth). We theoretically predict a width-dependent transition between depth-efficiency and depth-inefficiency in self-attention. We conduct systematic empirical ablations on networks of depths 6 to 48 that clearly reveal the theoretically predicted behaviors, and provide explicit quan"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2006.12467","kind":"arxiv","version":3},"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/2006.12467/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":"2006.12467","created_at":"2026-07-05T02:07:28.909368+00:00"},{"alias_kind":"arxiv_version","alias_value":"2006.12467v3","created_at":"2026-07-05T02:07:28.909368+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2006.12467","created_at":"2026-07-05T02:07:28.909368+00:00"},{"alias_kind":"pith_short_12","alias_value":"G2UAMCOBO7R7","created_at":"2026-07-05T02:07:28.909368+00:00"},{"alias_kind":"pith_short_16","alias_value":"G2UAMCOBO7R7B5PI","created_at":"2026-07-05T02:07:28.909368+00:00"},{"alias_kind":"pith_short_8","alias_value":"G2UAMCOB","created_at":"2026-07-05T02:07:28.909368+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.19944","citing_title":"A Measure-Theoretic Analysis of Reasoning: Structural Generalization and Approximation Limits","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2502.05171","citing_title":"Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach","ref_index":92,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08297","citing_title":"A Qualitative Test-Risk Mechanism for Scaling Behavior in Normalized Residual Networks","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2302.01318","citing_title":"Accelerating Large Language Model Decoding with Speculative Sampling","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2505.10465","citing_title":"Superposition Yields Robust Neural Scaling","ref_index":59,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/G2UAMCOBO7R7B5PI2FJJEZQNMZ","json":"https://pith.science/pith/G2UAMCOBO7R7B5PI2FJJEZQNMZ.json","graph_json":"https://pith.science/api/pith-number/G2UAMCOBO7R7B5PI2FJJEZQNMZ/graph.json","events_json":"https://pith.science/api/pith-number/G2UAMCOBO7R7B5PI2FJJEZQNMZ/events.json","paper":"https://pith.science/paper/G2UAMCOB"},"agent_actions":{"view_html":"https://pith.science/pith/G2UAMCOBO7R7B5PI2FJJEZQNMZ","download_json":"https://pith.science/pith/G2UAMCOBO7R7B5PI2FJJEZQNMZ.json","view_paper":"https://pith.science/paper/G2UAMCOB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2006.12467&json=true","fetch_graph":"https://pith.science/api/pith-number/G2UAMCOBO7R7B5PI2FJJEZQNMZ/graph.json","fetch_events":"https://pith.science/api/pith-number/G2UAMCOBO7R7B5PI2FJJEZQNMZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/G2UAMCOBO7R7B5PI2FJJEZQNMZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/G2UAMCOBO7R7B5PI2FJJEZQNMZ/action/storage_attestation","attest_author":"https://pith.science/pith/G2UAMCOBO7R7B5PI2FJJEZQNMZ/action/author_attestation","sign_citation":"https://pith.science/pith/G2UAMCOBO7R7B5PI2FJJEZQNMZ/action/citation_signature","submit_replication":"https://pith.science/pith/G2UAMCOBO7R7B5PI2FJJEZQNMZ/action/replication_record"}},"created_at":"2026-07-05T02:07:28.909368+00:00","updated_at":"2026-07-05T02:07:28.909368+00:00"}