{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:III6DKWD4HDYD2LZQULZA5KNHM","short_pith_number":"pith:III6DKWD","schema_version":"1.0","canonical_sha256":"4211e1aac3e1c781e979851790754d3b00573bbbe64335a7c4c56a1e39bbeeaf","source":{"kind":"arxiv","id":"2512.22473","version":5},"attestation_state":"computed","paper":{"title":"Gradient Dynamics of Attention: How Cross-Entropy Sculpts Bayesian Manifolds","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"stat.ML","authors_text":"Naman Agarwal, Siddhartha R. Dalal, Vishal Misra","submitted_at":"2025-12-27T05:31:44Z","abstract_excerpt":"Transformers empirically perform precise probabilistic reasoning in carefully constructed ``Bayesian wind tunnels'' and in large-scale language models, yet the mechanisms by which gradient-based learning creates the required internal geometry remain opaque. We provide a complete first-order analysis of how cross-entropy training reshapes attention scores and value vectors in a transformer attention head. Our core result is an \\emph{advantage-based routing law} for attention scores, \\[ \\frac{\\partial L}{\\partial s_{ij}} = \\alpha_{ij}\\bigl(b_{ij}-\\mathbb{E}_{\\alpha_i}[b]\\bigr), \\qquad b_{ij} := "},"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":"2512.22473","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2025-12-27T05:31:44Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"037974fe4d2a4a2ba25aaec11e9e7b7487504ea8d9dba8845122028211d8539c","abstract_canon_sha256":"d1ca30409b9aae2824a871c4c275fb14d12420c66be5c83ba13e48dc1a054563"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:03:00.933799Z","signature_b64":"DDh1srDpzYn40QcZOayobSF2GmIn1rXKwEk4v1V7z0uNsUYzEH5v2C7dumtE0Fwk371elHJ4ijUvUQJmrU1NCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4211e1aac3e1c781e979851790754d3b00573bbbe64335a7c4c56a1e39bbeeaf","last_reissued_at":"2026-05-20T00:03:00.932904Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:03:00.932904Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Gradient Dynamics of Attention: How Cross-Entropy Sculpts Bayesian Manifolds","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"stat.ML","authors_text":"Naman Agarwal, Siddhartha R. Dalal, Vishal Misra","submitted_at":"2025-12-27T05:31:44Z","abstract_excerpt":"Transformers empirically perform precise probabilistic reasoning in carefully constructed ``Bayesian wind tunnels'' and in large-scale language models, yet the mechanisms by which gradient-based learning creates the required internal geometry remain opaque. We provide a complete first-order analysis of how cross-entropy training reshapes attention scores and value vectors in a transformer attention head. Our core result is an \\emph{advantage-based routing law} for attention scores, \\[ \\frac{\\partial L}{\\partial s_{ij}} = \\alpha_{ij}\\bigl(b_{ij}-\\mathbb{E}_{\\alpha_i}[b]\\bigr), \\qquad b_{ij} := "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2512.22473","kind":"arxiv","version":5},"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/2512.22473/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":"2512.22473","created_at":"2026-05-20T00:03:00.933083+00:00"},{"alias_kind":"arxiv_version","alias_value":"2512.22473v5","created_at":"2026-05-20T00:03:00.933083+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2512.22473","created_at":"2026-05-20T00:03:00.933083+00:00"},{"alias_kind":"pith_short_12","alias_value":"III6DKWD4HDY","created_at":"2026-05-20T00:03:00.933083+00:00"},{"alias_kind":"pith_short_16","alias_value":"III6DKWD4HDYD2LZ","created_at":"2026-05-20T00:03:00.933083+00:00"},{"alias_kind":"pith_short_8","alias_value":"III6DKWD","created_at":"2026-05-20T00:03:00.933083+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2512.22471","citing_title":"The Bayesian Geometry of Transformer Attention","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2512.23752","citing_title":"Geometric Scaling of Bayesian Inference in LLMs","ref_index":1,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/III6DKWD4HDYD2LZQULZA5KNHM","json":"https://pith.science/pith/III6DKWD4HDYD2LZQULZA5KNHM.json","graph_json":"https://pith.science/api/pith-number/III6DKWD4HDYD2LZQULZA5KNHM/graph.json","events_json":"https://pith.science/api/pith-number/III6DKWD4HDYD2LZQULZA5KNHM/events.json","paper":"https://pith.science/paper/III6DKWD"},"agent_actions":{"view_html":"https://pith.science/pith/III6DKWD4HDYD2LZQULZA5KNHM","download_json":"https://pith.science/pith/III6DKWD4HDYD2LZQULZA5KNHM.json","view_paper":"https://pith.science/paper/III6DKWD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2512.22473&json=true","fetch_graph":"https://pith.science/api/pith-number/III6DKWD4HDYD2LZQULZA5KNHM/graph.json","fetch_events":"https://pith.science/api/pith-number/III6DKWD4HDYD2LZQULZA5KNHM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/III6DKWD4HDYD2LZQULZA5KNHM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/III6DKWD4HDYD2LZQULZA5KNHM/action/storage_attestation","attest_author":"https://pith.science/pith/III6DKWD4HDYD2LZQULZA5KNHM/action/author_attestation","sign_citation":"https://pith.science/pith/III6DKWD4HDYD2LZQULZA5KNHM/action/citation_signature","submit_replication":"https://pith.science/pith/III6DKWD4HDYD2LZQULZA5KNHM/action/replication_record"}},"created_at":"2026-05-20T00:03:00.933083+00:00","updated_at":"2026-05-20T00:03:00.933083+00:00"}