{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:57OFR37VUGR2LT25XAXWQP6ANQ","short_pith_number":"pith:57OFR37V","schema_version":"1.0","canonical_sha256":"efdc58eff5a1a3a5cf5db82f683fc06c15df63955a2c6b3411d48789e99a2c50","source":{"kind":"arxiv","id":"2403.02579","version":1},"attestation_state":"computed","paper":{"title":"Geometric Dynamics of Signal Propagation Predict Trainability of Transformers","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cond-mat.dis-nn","authors_text":"Aditya Cowsik, Surya Ganguli, Tamra Nebabu, Xiao-Liang Qi","submitted_at":"2024-03-05T01:30:34Z","abstract_excerpt":"We investigate forward signal propagation and gradient back propagation in deep, randomly initialized transformers, yielding simple necessary and sufficient conditions on initialization hyperparameters that ensure trainability of deep transformers. Our approach treats the evolution of the representations of $n$ tokens as they propagate through the transformer layers in terms of a discrete time dynamical system of $n$ interacting particles. We derive simple update equations for the evolving geometry of this particle system, starting from a permutation symmetric simplex. Our update equations sho"},"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":"2403.02579","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cond-mat.dis-nn","submitted_at":"2024-03-05T01:30:34Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"e1eb54b47da5b2ea1ae41e5c12f1e05bd904c7c2af09afe067539add86f3e767","abstract_canon_sha256":"3462621975d18470d7e48bd99b7136a9f506e194b6cc9ad4cc4b13c8341b35dd"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:52:19.265139Z","signature_b64":"NCs7Z9RMQAhIswvzsDaAK/oS3fLCZhMz55xtnlFw7Kxv9Ne2L+8W+HnGcYK+kZfKiKGEz4AjN/9m6MFZLIc2Cw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"efdc58eff5a1a3a5cf5db82f683fc06c15df63955a2c6b3411d48789e99a2c50","last_reissued_at":"2026-07-05T07:52:19.264630Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:52:19.264630Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Geometric Dynamics of Signal Propagation Predict Trainability of Transformers","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cond-mat.dis-nn","authors_text":"Aditya Cowsik, Surya Ganguli, Tamra Nebabu, Xiao-Liang Qi","submitted_at":"2024-03-05T01:30:34Z","abstract_excerpt":"We investigate forward signal propagation and gradient back propagation in deep, randomly initialized transformers, yielding simple necessary and sufficient conditions on initialization hyperparameters that ensure trainability of deep transformers. Our approach treats the evolution of the representations of $n$ tokens as they propagate through the transformer layers in terms of a discrete time dynamical system of $n$ interacting particles. We derive simple update equations for the evolving geometry of this particle system, starting from a permutation symmetric simplex. Our update equations sho"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.02579","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/2403.02579/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":"2403.02579","created_at":"2026-07-05T07:52:19.264689+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.02579v1","created_at":"2026-07-05T07:52:19.264689+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.02579","created_at":"2026-07-05T07:52:19.264689+00:00"},{"alias_kind":"pith_short_12","alias_value":"57OFR37VUGR2","created_at":"2026-07-05T07:52:19.264689+00:00"},{"alias_kind":"pith_short_16","alias_value":"57OFR37VUGR2LT25","created_at":"2026-07-05T07:52:19.264689+00:00"},{"alias_kind":"pith_short_8","alias_value":"57OFR37V","created_at":"2026-07-05T07:52:19.264689+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.13300","citing_title":"Quantizing Time-Series Models As Dynamical Systems: Trajectory-Based Quantization Sensitivity Score","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2505.24333","citing_title":"Two failure modes of deep transformers and how to avoid them: a unified theory of signal propagation at initialisation","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12697","citing_title":"A Unified Framework for Critical Scaling of Inverse Temperature in Self-Attention","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/57OFR37VUGR2LT25XAXWQP6ANQ","json":"https://pith.science/pith/57OFR37VUGR2LT25XAXWQP6ANQ.json","graph_json":"https://pith.science/api/pith-number/57OFR37VUGR2LT25XAXWQP6ANQ/graph.json","events_json":"https://pith.science/api/pith-number/57OFR37VUGR2LT25XAXWQP6ANQ/events.json","paper":"https://pith.science/paper/57OFR37V"},"agent_actions":{"view_html":"https://pith.science/pith/57OFR37VUGR2LT25XAXWQP6ANQ","download_json":"https://pith.science/pith/57OFR37VUGR2LT25XAXWQP6ANQ.json","view_paper":"https://pith.science/paper/57OFR37V","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.02579&json=true","fetch_graph":"https://pith.science/api/pith-number/57OFR37VUGR2LT25XAXWQP6ANQ/graph.json","fetch_events":"https://pith.science/api/pith-number/57OFR37VUGR2LT25XAXWQP6ANQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/57OFR37VUGR2LT25XAXWQP6ANQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/57OFR37VUGR2LT25XAXWQP6ANQ/action/storage_attestation","attest_author":"https://pith.science/pith/57OFR37VUGR2LT25XAXWQP6ANQ/action/author_attestation","sign_citation":"https://pith.science/pith/57OFR37VUGR2LT25XAXWQP6ANQ/action/citation_signature","submit_replication":"https://pith.science/pith/57OFR37VUGR2LT25XAXWQP6ANQ/action/replication_record"}},"created_at":"2026-07-05T07:52:19.264689+00:00","updated_at":"2026-07-05T07:52:19.264689+00:00"}