{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:XXWNFJEDZ666ERFBEXT5EOC53I","short_pith_number":"pith:XXWNFJED","schema_version":"1.0","canonical_sha256":"bdecd2a483cfbde244a125e7d2385dda387e5e8efd3a59c7cc0334c29b893edd","source":{"kind":"arxiv","id":"2007.12927","version":4},"attestation_state":"computed","paper":{"title":"Neural networks with late-phase weights","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV","stat.ML"],"primary_cat":"cs.LG","authors_text":"Alexander Meulemans, Benjamin F. Grewe, Christian Henning, Jo\\~ao Sacramento, Johannes von Oswald, Seijin Kobayashi","submitted_at":"2020-07-25T13:23:37Z","abstract_excerpt":"The largely successful method of training neural networks is to learn their weights using some variant of stochastic gradient descent (SGD). Here, we show that the solutions found by SGD can be further improved by ensembling a subset of the weights in late stages of learning. At the end of learning, we obtain back a single model by taking a spatial average in weight space. To avoid incurring increased computational costs, we investigate a family of low-dimensional late-phase weight models which interact multiplicatively with the remaining parameters. Our results show that augmenting standard m"},"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":"2007.12927","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2020-07-25T13:23:37Z","cross_cats_sorted":["cs.CV","stat.ML"],"title_canon_sha256":"3b2c5bd1e25bdef3082fd47a30df9d17d2893bbaad848c80d82306e7ef986554","abstract_canon_sha256":"0cbb94da934f22d5a85ba2eae5c07e9665e25f8b7f9bd64a3d9837601e2d8847"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:13:08.567884Z","signature_b64":"6SmueiPhxeF5dN4IenJ/sB3MSkQdyY/dPj2YGAwao75hewblJgWaKEKrU3pmPHE9ROKtNa/zIO15wWqCi4pbCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bdecd2a483cfbde244a125e7d2385dda387e5e8efd3a59c7cc0334c29b893edd","last_reissued_at":"2026-07-05T04:13:08.567457Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:13:08.567457Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Neural networks with late-phase weights","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV","stat.ML"],"primary_cat":"cs.LG","authors_text":"Alexander Meulemans, Benjamin F. Grewe, Christian Henning, Jo\\~ao Sacramento, Johannes von Oswald, Seijin Kobayashi","submitted_at":"2020-07-25T13:23:37Z","abstract_excerpt":"The largely successful method of training neural networks is to learn their weights using some variant of stochastic gradient descent (SGD). Here, we show that the solutions found by SGD can be further improved by ensembling a subset of the weights in late stages of learning. At the end of learning, we obtain back a single model by taking a spatial average in weight space. To avoid incurring increased computational costs, we investigate a family of low-dimensional late-phase weight models which interact multiplicatively with the remaining parameters. Our results show that augmenting standard m"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2007.12927","kind":"arxiv","version":4},"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/2007.12927/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":"2007.12927","created_at":"2026-07-05T04:13:08.567514+00:00"},{"alias_kind":"arxiv_version","alias_value":"2007.12927v4","created_at":"2026-07-05T04:13:08.567514+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2007.12927","created_at":"2026-07-05T04:13:08.567514+00:00"},{"alias_kind":"pith_short_12","alias_value":"XXWNFJEDZ666","created_at":"2026-07-05T04:13:08.567514+00:00"},{"alias_kind":"pith_short_16","alias_value":"XXWNFJEDZ666ERFB","created_at":"2026-07-05T04:13:08.567514+00:00"},{"alias_kind":"pith_short_8","alias_value":"XXWNFJED","created_at":"2026-07-05T04:13:08.567514+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25087","citing_title":"Neural Network Quantization by Learning Low-Loss Subspaces","ref_index":46,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XXWNFJEDZ666ERFBEXT5EOC53I","json":"https://pith.science/pith/XXWNFJEDZ666ERFBEXT5EOC53I.json","graph_json":"https://pith.science/api/pith-number/XXWNFJEDZ666ERFBEXT5EOC53I/graph.json","events_json":"https://pith.science/api/pith-number/XXWNFJEDZ666ERFBEXT5EOC53I/events.json","paper":"https://pith.science/paper/XXWNFJED"},"agent_actions":{"view_html":"https://pith.science/pith/XXWNFJEDZ666ERFBEXT5EOC53I","download_json":"https://pith.science/pith/XXWNFJEDZ666ERFBEXT5EOC53I.json","view_paper":"https://pith.science/paper/XXWNFJED","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2007.12927&json=true","fetch_graph":"https://pith.science/api/pith-number/XXWNFJEDZ666ERFBEXT5EOC53I/graph.json","fetch_events":"https://pith.science/api/pith-number/XXWNFJEDZ666ERFBEXT5EOC53I/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XXWNFJEDZ666ERFBEXT5EOC53I/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XXWNFJEDZ666ERFBEXT5EOC53I/action/storage_attestation","attest_author":"https://pith.science/pith/XXWNFJEDZ666ERFBEXT5EOC53I/action/author_attestation","sign_citation":"https://pith.science/pith/XXWNFJEDZ666ERFBEXT5EOC53I/action/citation_signature","submit_replication":"https://pith.science/pith/XXWNFJEDZ666ERFBEXT5EOC53I/action/replication_record"}},"created_at":"2026-07-05T04:13:08.567514+00:00","updated_at":"2026-07-05T04:13:08.567514+00:00"}