{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:3S6C6YCTNTKCU2KG5YC2K7QIQC","short_pith_number":"pith:3S6C6YCT","schema_version":"1.0","canonical_sha256":"dcbc2f60536cd42a6946ee05a57e088082a50461aed0a2918e1dcfc8134cf8db","source":{"kind":"arxiv","id":"2402.09268","version":1},"attestation_state":"computed","paper":{"title":"Transformers, parallel computation, and logarithmic depth","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Clayton Sanford, Daniel Hsu, Matus Telgarsky","submitted_at":"2024-02-14T15:54:55Z","abstract_excerpt":"We show that a constant number of self-attention layers can efficiently simulate, and be simulated by, a constant number of communication rounds of Massively Parallel Computation. As a consequence, we show that logarithmic depth is sufficient for transformers to solve basic computational tasks that cannot be efficiently solved by several other neural sequence models and sub-quadratic transformer approximations. We thus establish parallelism as a key distinguishing property of transformers."},"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":"2402.09268","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-14T15:54:55Z","cross_cats_sorted":[],"title_canon_sha256":"36c32ba93809ecc3193bed003dc5aee9afcd485e899cda3fd666214576d4e5d0","abstract_canon_sha256":"40add66b1a374f9dca4da109b950c5c86582dc6d39beb991b75686c23be89b0f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:45:11.008659Z","signature_b64":"Q7nP/+HYs0jpVfdznX0deAZPkSDKXw5aXKfFyXOYi67tGnu65tmI0kHpncM7LAVMEOQc5WyKBrISFyGvDDLoAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dcbc2f60536cd42a6946ee05a57e088082a50461aed0a2918e1dcfc8134cf8db","last_reissued_at":"2026-07-05T07:45:11.008217Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:45:11.008217Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Transformers, parallel computation, and logarithmic depth","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Clayton Sanford, Daniel Hsu, Matus Telgarsky","submitted_at":"2024-02-14T15:54:55Z","abstract_excerpt":"We show that a constant number of self-attention layers can efficiently simulate, and be simulated by, a constant number of communication rounds of Massively Parallel Computation. As a consequence, we show that logarithmic depth is sufficient for transformers to solve basic computational tasks that cannot be efficiently solved by several other neural sequence models and sub-quadratic transformer approximations. We thus establish parallelism as a key distinguishing property of transformers."},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.09268","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/2402.09268/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":"2402.09268","created_at":"2026-07-05T07:45:11.008278+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.09268v1","created_at":"2026-07-05T07:45:11.008278+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.09268","created_at":"2026-07-05T07:45:11.008278+00:00"},{"alias_kind":"pith_short_12","alias_value":"3S6C6YCTNTKC","created_at":"2026-07-05T07:45:11.008278+00:00"},{"alias_kind":"pith_short_16","alias_value":"3S6C6YCTNTKCU2KG","created_at":"2026-07-05T07:45:11.008278+00:00"},{"alias_kind":"pith_short_8","alias_value":"3S6C6YCT","created_at":"2026-07-05T07:45:11.008278+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.11673","citing_title":"Higher-Order Token Interactions via Quantum Attention","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22471","citing_title":"Lost in Tokenization: Fundamental Trade-offs in Graph Tokenization for Transformers","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2510.25741","citing_title":"Scaling Latent Reasoning via Looped Language Models","ref_index":79,"is_internal_anchor":false},{"citing_arxiv_id":"2603.26554","citing_title":"Sharp Capacity Scaling of Spectral Optimizers in Learning Associative Memory","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24878","citing_title":"Transformer Approximations from ReLUs","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21691","citing_title":"There Will Be a Scientific Theory of Deep Learning","ref_index":245,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17935","citing_title":"How Much Cache Does Reasoning Need? Depth-Cache Tradeoffs in KV-Compressed Transformers","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3S6C6YCTNTKCU2KG5YC2K7QIQC","json":"https://pith.science/pith/3S6C6YCTNTKCU2KG5YC2K7QIQC.json","graph_json":"https://pith.science/api/pith-number/3S6C6YCTNTKCU2KG5YC2K7QIQC/graph.json","events_json":"https://pith.science/api/pith-number/3S6C6YCTNTKCU2KG5YC2K7QIQC/events.json","paper":"https://pith.science/paper/3S6C6YCT"},"agent_actions":{"view_html":"https://pith.science/pith/3S6C6YCTNTKCU2KG5YC2K7QIQC","download_json":"https://pith.science/pith/3S6C6YCTNTKCU2KG5YC2K7QIQC.json","view_paper":"https://pith.science/paper/3S6C6YCT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.09268&json=true","fetch_graph":"https://pith.science/api/pith-number/3S6C6YCTNTKCU2KG5YC2K7QIQC/graph.json","fetch_events":"https://pith.science/api/pith-number/3S6C6YCTNTKCU2KG5YC2K7QIQC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3S6C6YCTNTKCU2KG5YC2K7QIQC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3S6C6YCTNTKCU2KG5YC2K7QIQC/action/storage_attestation","attest_author":"https://pith.science/pith/3S6C6YCTNTKCU2KG5YC2K7QIQC/action/author_attestation","sign_citation":"https://pith.science/pith/3S6C6YCTNTKCU2KG5YC2K7QIQC/action/citation_signature","submit_replication":"https://pith.science/pith/3S6C6YCTNTKCU2KG5YC2K7QIQC/action/replication_record"}},"created_at":"2026-07-05T07:45:11.008278+00:00","updated_at":"2026-07-05T07:45:11.008278+00:00"}