{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:ZUUMESTNS7HDMCJWR5TS4OQOFU","short_pith_number":"pith:ZUUMESTN","schema_version":"1.0","canonical_sha256":"cd28c24a6d97ce3609368f672e3a0e2d26fc19f33d8934a4f1c2a66d160be6f9","source":{"kind":"arxiv","id":"1912.00018","version":1},"attestation_state":"computed","paper":{"title":"On the Heavy-Tailed Theory of Stochastic Gradient Descent for Deep Neural Networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG","math.CA"],"primary_cat":"stat.ML","authors_text":"Ga\\\"el Richard, Levent Sagun, Mert G\\\"urb\\\"uzbalaban, Thanh Huy Nguyen, Umut \\c{S}im\\c{s}ekli","submitted_at":"2019-11-29T16:56:02Z","abstract_excerpt":"The gradient noise (GN) in the stochastic gradient descent (SGD) algorithm is often considered to be Gaussian in the large data regime by assuming that the \\emph{classical} central limit theorem (CLT) kicks in. This assumption is often made for mathematical convenience, since it enables SGD to be analyzed as a stochastic differential equation (SDE) driven by a Brownian motion. We argue that the Gaussianity assumption might fail to hold in deep learning settings and hence render the Brownian motion-based analyses inappropriate. Inspired by non-Gaussian natural phenomena, we consider the GN in a"},"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":"1912.00018","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2019-11-29T16:56:02Z","cross_cats_sorted":["cs.LG","math.CA"],"title_canon_sha256":"3af76bde47635e27cc1cee567bcdb31bb9d6722e8ea45bfc38aff7e8d922c3bc","abstract_canon_sha256":"9c3d52bbce9be56c4ba6c3c37ef244a2668b2e8aba158007b6b64d6e2dec87ac"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T00:22:54.290833Z","signature_b64":"HmkWpuRBuoiT9HQQRByjS2+hxZDtfVFE4v8a8dmmxJziEPqT7CD01ibtooskfwVYVnb4dLL31HKPnPZ0n8VjBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cd28c24a6d97ce3609368f672e3a0e2d26fc19f33d8934a4f1c2a66d160be6f9","last_reissued_at":"2026-07-05T00:22:54.290453Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T00:22:54.290453Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On the Heavy-Tailed Theory of Stochastic Gradient Descent for Deep Neural Networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG","math.CA"],"primary_cat":"stat.ML","authors_text":"Ga\\\"el Richard, Levent Sagun, Mert G\\\"urb\\\"uzbalaban, Thanh Huy Nguyen, Umut \\c{S}im\\c{s}ekli","submitted_at":"2019-11-29T16:56:02Z","abstract_excerpt":"The gradient noise (GN) in the stochastic gradient descent (SGD) algorithm is often considered to be Gaussian in the large data regime by assuming that the \\emph{classical} central limit theorem (CLT) kicks in. This assumption is often made for mathematical convenience, since it enables SGD to be analyzed as a stochastic differential equation (SDE) driven by a Brownian motion. We argue that the Gaussianity assumption might fail to hold in deep learning settings and hence render the Brownian motion-based analyses inappropriate. Inspired by non-Gaussian natural phenomena, we consider the GN in a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1912.00018","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/1912.00018/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":"1912.00018","created_at":"2026-07-05T00:22:54.290508+00:00"},{"alias_kind":"arxiv_version","alias_value":"1912.00018v1","created_at":"2026-07-05T00:22:54.290508+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1912.00018","created_at":"2026-07-05T00:22:54.290508+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZUUMESTNS7HD","created_at":"2026-07-05T00:22:54.290508+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZUUMESTNS7HDMCJW","created_at":"2026-07-05T00:22:54.290508+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZUUMESTN","created_at":"2026-07-05T00:22:54.290508+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.26000","citing_title":"Statistical Inference for Stochastic Gradient Descent: Beyond Finite Variance","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07184","citing_title":"Convergence of Stochastic Gradient Descent with mini-batching and infinite variance","ref_index":6,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZUUMESTNS7HDMCJWR5TS4OQOFU","json":"https://pith.science/pith/ZUUMESTNS7HDMCJWR5TS4OQOFU.json","graph_json":"https://pith.science/api/pith-number/ZUUMESTNS7HDMCJWR5TS4OQOFU/graph.json","events_json":"https://pith.science/api/pith-number/ZUUMESTNS7HDMCJWR5TS4OQOFU/events.json","paper":"https://pith.science/paper/ZUUMESTN"},"agent_actions":{"view_html":"https://pith.science/pith/ZUUMESTNS7HDMCJWR5TS4OQOFU","download_json":"https://pith.science/pith/ZUUMESTNS7HDMCJWR5TS4OQOFU.json","view_paper":"https://pith.science/paper/ZUUMESTN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1912.00018&json=true","fetch_graph":"https://pith.science/api/pith-number/ZUUMESTNS7HDMCJWR5TS4OQOFU/graph.json","fetch_events":"https://pith.science/api/pith-number/ZUUMESTNS7HDMCJWR5TS4OQOFU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZUUMESTNS7HDMCJWR5TS4OQOFU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZUUMESTNS7HDMCJWR5TS4OQOFU/action/storage_attestation","attest_author":"https://pith.science/pith/ZUUMESTNS7HDMCJWR5TS4OQOFU/action/author_attestation","sign_citation":"https://pith.science/pith/ZUUMESTNS7HDMCJWR5TS4OQOFU/action/citation_signature","submit_replication":"https://pith.science/pith/ZUUMESTNS7HDMCJWR5TS4OQOFU/action/replication_record"}},"created_at":"2026-07-05T00:22:54.290508+00:00","updated_at":"2026-07-05T00:22:54.290508+00:00"}