{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:2NY7NDPJVWS6G4Y4NCSLEVXJXX","short_pith_number":"pith:2NY7NDPJ","schema_version":"1.0","canonical_sha256":"d371f68de9ada5e3731c68a4b256e9bdcd58c99dd37b2d04deddaaaeacf1aa77","source":{"kind":"arxiv","id":"2203.11086","version":2},"attestation_state":"computed","paper":{"title":"Overcoming Oscillations in Quantization-Aware Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Marios Fournarakis, Markus Nagel, Tijmen Blankevoort, Yelysei Bondarenko","submitted_at":"2022-03-21T16:07:42Z","abstract_excerpt":"When training neural networks with simulated quantization, we observe that quantized weights can, rather unexpectedly, oscillate between two grid-points. The importance of this effect and its impact on quantization-aware training (QAT) are not well-understood or investigated in literature. In this paper, we delve deeper into the phenomenon of weight oscillations and show that it can lead to a significant accuracy degradation due to wrongly estimated batch-normalization statistics during inference and increased noise during training. These effects are particularly pronounced in low-bit ($\\leq$ "},"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":"2203.11086","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2022-03-21T16:07:42Z","cross_cats_sorted":[],"title_canon_sha256":"837ed527ad635352f8971eac10893c682f9c8a578c77e1bb49339bf3736a7f37","abstract_canon_sha256":"0f156592d674bfed5650a8f74d9de70f119096b00ab27497a24663d3ed0cb100"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:35:59.675991Z","signature_b64":"IVEkBxV6ciakacbAgneHXxuqDb8qJV5A/tH/KdMnjsN65xyy1DOXZr73zzaOn3m1rwrD/HQbSdwaIfUBbt5aBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d371f68de9ada5e3731c68a4b256e9bdcd58c99dd37b2d04deddaaaeacf1aa77","last_reissued_at":"2026-07-05T04:35:59.675491Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:35:59.675491Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Overcoming Oscillations in Quantization-Aware Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Marios Fournarakis, Markus Nagel, Tijmen Blankevoort, Yelysei Bondarenko","submitted_at":"2022-03-21T16:07:42Z","abstract_excerpt":"When training neural networks with simulated quantization, we observe that quantized weights can, rather unexpectedly, oscillate between two grid-points. The importance of this effect and its impact on quantization-aware training (QAT) are not well-understood or investigated in literature. In this paper, we delve deeper into the phenomenon of weight oscillations and show that it can lead to a significant accuracy degradation due to wrongly estimated batch-normalization statistics during inference and increased noise during training. These effects are particularly pronounced in low-bit ($\\leq$ "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2203.11086","kind":"arxiv","version":2},"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/2203.11086/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":"2203.11086","created_at":"2026-07-05T04:35:59.675549+00:00"},{"alias_kind":"arxiv_version","alias_value":"2203.11086v2","created_at":"2026-07-05T04:35:59.675549+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2203.11086","created_at":"2026-07-05T04:35:59.675549+00:00"},{"alias_kind":"pith_short_12","alias_value":"2NY7NDPJVWS6","created_at":"2026-07-05T04:35:59.675549+00:00"},{"alias_kind":"pith_short_16","alias_value":"2NY7NDPJVWS6G4Y4","created_at":"2026-07-05T04:35:59.675549+00:00"},{"alias_kind":"pith_short_8","alias_value":"2NY7NDPJ","created_at":"2026-07-05T04:35:59.675549+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.25966","citing_title":"Mapping the Schedule x Bit-Width Boundary in Sub-100M Quantisation-Aware Training","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2606.27759","citing_title":"Layerwise Progressive Freezing: A Training Scaffold for Depth-Scalable Binary Networks","ref_index":41,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2NY7NDPJVWS6G4Y4NCSLEVXJXX","json":"https://pith.science/pith/2NY7NDPJVWS6G4Y4NCSLEVXJXX.json","graph_json":"https://pith.science/api/pith-number/2NY7NDPJVWS6G4Y4NCSLEVXJXX/graph.json","events_json":"https://pith.science/api/pith-number/2NY7NDPJVWS6G4Y4NCSLEVXJXX/events.json","paper":"https://pith.science/paper/2NY7NDPJ"},"agent_actions":{"view_html":"https://pith.science/pith/2NY7NDPJVWS6G4Y4NCSLEVXJXX","download_json":"https://pith.science/pith/2NY7NDPJVWS6G4Y4NCSLEVXJXX.json","view_paper":"https://pith.science/paper/2NY7NDPJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2203.11086&json=true","fetch_graph":"https://pith.science/api/pith-number/2NY7NDPJVWS6G4Y4NCSLEVXJXX/graph.json","fetch_events":"https://pith.science/api/pith-number/2NY7NDPJVWS6G4Y4NCSLEVXJXX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2NY7NDPJVWS6G4Y4NCSLEVXJXX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2NY7NDPJVWS6G4Y4NCSLEVXJXX/action/storage_attestation","attest_author":"https://pith.science/pith/2NY7NDPJVWS6G4Y4NCSLEVXJXX/action/author_attestation","sign_citation":"https://pith.science/pith/2NY7NDPJVWS6G4Y4NCSLEVXJXX/action/citation_signature","submit_replication":"https://pith.science/pith/2NY7NDPJVWS6G4Y4NCSLEVXJXX/action/replication_record"}},"created_at":"2026-07-05T04:35:59.675549+00:00","updated_at":"2026-07-05T04:35:59.675549+00:00"}