{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:XF5I7QWDJIMATFESUGE3MRLOEM","short_pith_number":"pith:XF5I7QWD","schema_version":"1.0","canonical_sha256":"b97a8fc2c34a18099492a189b6456e2302b2a6a0561dc245f21901f03ea0adcb","source":{"kind":"arxiv","id":"2502.05668","version":3},"attestation_state":"computed","paper":{"title":"The late-stage training dynamics of (stochastic) subgradient descent on homogeneous neural networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.NE","math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Nicolas Schreuder, Sholom Schechtman","submitted_at":"2025-02-08T19:09:16Z","abstract_excerpt":"We analyze the implicit bias of constant step stochastic subgradient descent (SGD). We consider the setting of binary classification with homogeneous neural networks - a large class of deep neural networks with ReLU-type activation functions such as MLPs and CNNs without biases. We interpret the dynamics of normalized SGD iterates as an Euler-like discretization of a conservative field flow that is naturally associated to the normalized classification margin. Owing to this interpretation, we show that normalized SGD iterates converge to the set of critical points of the normalized margin at la"},"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":"2502.05668","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-02-08T19:09:16Z","cross_cats_sorted":["cs.NE","math.OC","stat.ML"],"title_canon_sha256":"17149e74f1b9a141677080f2be380b533c7217b483276080d06ca974308b0c4f","abstract_canon_sha256":"b38cce3f3c393cdb63a410bea95168671760bbc5ea5cb698473b459b7278a96d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:38:50.149434Z","signature_b64":"ASaulEUPqSZyAOhnTdniSSucpmzim9iSnyY7K/1ncYOA/HSqyzzrd3TwQJlya/k7E27qmZfHTY9/+9PEAGVrCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b97a8fc2c34a18099492a189b6456e2302b2a6a0561dc245f21901f03ea0adcb","last_reissued_at":"2026-07-05T11:38:50.148995Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:38:50.148995Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The late-stage training dynamics of (stochastic) subgradient descent on homogeneous neural networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.NE","math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Nicolas Schreuder, Sholom Schechtman","submitted_at":"2025-02-08T19:09:16Z","abstract_excerpt":"We analyze the implicit bias of constant step stochastic subgradient descent (SGD). We consider the setting of binary classification with homogeneous neural networks - a large class of deep neural networks with ReLU-type activation functions such as MLPs and CNNs without biases. We interpret the dynamics of normalized SGD iterates as an Euler-like discretization of a conservative field flow that is naturally associated to the normalized classification margin. Owing to this interpretation, we show that normalized SGD iterates converge to the set of critical points of the normalized margin at la"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.05668","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/2502.05668/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":"2502.05668","created_at":"2026-07-05T11:38:50.149054+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.05668v3","created_at":"2026-07-05T11:38:50.149054+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.05668","created_at":"2026-07-05T11:38:50.149054+00:00"},{"alias_kind":"pith_short_12","alias_value":"XF5I7QWDJIMA","created_at":"2026-07-05T11:38:50.149054+00:00"},{"alias_kind":"pith_short_16","alias_value":"XF5I7QWDJIMATFES","created_at":"2026-07-05T11:38:50.149054+00:00"},{"alias_kind":"pith_short_8","alias_value":"XF5I7QWD","created_at":"2026-07-05T11:38:50.149054+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.30559","citing_title":"Convergence of Continual Learning in Homogeneous Deep Networks","ref_index":8,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XF5I7QWDJIMATFESUGE3MRLOEM","json":"https://pith.science/pith/XF5I7QWDJIMATFESUGE3MRLOEM.json","graph_json":"https://pith.science/api/pith-number/XF5I7QWDJIMATFESUGE3MRLOEM/graph.json","events_json":"https://pith.science/api/pith-number/XF5I7QWDJIMATFESUGE3MRLOEM/events.json","paper":"https://pith.science/paper/XF5I7QWD"},"agent_actions":{"view_html":"https://pith.science/pith/XF5I7QWDJIMATFESUGE3MRLOEM","download_json":"https://pith.science/pith/XF5I7QWDJIMATFESUGE3MRLOEM.json","view_paper":"https://pith.science/paper/XF5I7QWD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.05668&json=true","fetch_graph":"https://pith.science/api/pith-number/XF5I7QWDJIMATFESUGE3MRLOEM/graph.json","fetch_events":"https://pith.science/api/pith-number/XF5I7QWDJIMATFESUGE3MRLOEM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XF5I7QWDJIMATFESUGE3MRLOEM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XF5I7QWDJIMATFESUGE3MRLOEM/action/storage_attestation","attest_author":"https://pith.science/pith/XF5I7QWDJIMATFESUGE3MRLOEM/action/author_attestation","sign_citation":"https://pith.science/pith/XF5I7QWDJIMATFESUGE3MRLOEM/action/citation_signature","submit_replication":"https://pith.science/pith/XF5I7QWDJIMATFESUGE3MRLOEM/action/replication_record"}},"created_at":"2026-07-05T11:38:50.149054+00:00","updated_at":"2026-07-05T11:38:50.149054+00:00"}