{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:U72GGMCAE7J3GMPF7W3AYPXJDU","short_pith_number":"pith:U72GGMCA","schema_version":"1.0","canonical_sha256":"a7f463304027d3b331e5fdb60c3ee91d3d71acac05d225a2d37a5df64231f2d8","source":{"kind":"arxiv","id":"2510.18784","version":3},"attestation_state":"computed","paper":{"title":"CAGE: Curvature-Aware Gradient Estimation For Accurate Quantization-Aware Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alexandra Volkova, Andrei Panferov, Dan Alistarh, Mher Safaryan, Soroush Tabesh","submitted_at":"2025-10-21T16:33:57Z","abstract_excerpt":"Despite significant work on low-bit quantization-aware training (QAT), there is still an accuracy gap between such techniques and native training. To address this, we introduce CAGE (Curvature-Aware Gradient Estimation), a new QAT method that augments the straight-through estimator (STE) gradient with a curvature-aware correction designed to counteract the loss increase induced by quantization. CAGE is derived from a multi-objective view of QAT that balances loss minimization with the quantization constraints, yielding a principled correction term that depends on local curvature information. O"},"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":"2510.18784","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-10-21T16:33:57Z","cross_cats_sorted":[],"title_canon_sha256":"8699cae96978819ed61b29bb7fc994f0f3c7ef9113da0a2dfa532afe0b1a9647","abstract_canon_sha256":"6addfec1d659969ccf3815fd54578cdd083b3ba416a901fb4fe800b4a4ddc083"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-19T16:12:48.673344Z","signature_b64":"1Y85GUb06KNF2NBDYgAxwIOURTAOtn0TRQ5n7x+cUCPOB82wC6FTr1q2cJy2Dsl4D2UO1VauNEBmgDyJ7Y6rDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a7f463304027d3b331e5fdb60c3ee91d3d71acac05d225a2d37a5df64231f2d8","last_reissued_at":"2026-06-19T16:12:48.672864Z","signature_status":"signed_v1","first_computed_at":"2026-06-19T16:12:48.672864Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"CAGE: Curvature-Aware Gradient Estimation For Accurate Quantization-Aware Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alexandra Volkova, Andrei Panferov, Dan Alistarh, Mher Safaryan, Soroush Tabesh","submitted_at":"2025-10-21T16:33:57Z","abstract_excerpt":"Despite significant work on low-bit quantization-aware training (QAT), there is still an accuracy gap between such techniques and native training. To address this, we introduce CAGE (Curvature-Aware Gradient Estimation), a new QAT method that augments the straight-through estimator (STE) gradient with a curvature-aware correction designed to counteract the loss increase induced by quantization. CAGE is derived from a multi-objective view of QAT that balances loss minimization with the quantization constraints, yielding a principled correction term that depends on local curvature information. O"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2510.18784","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/2510.18784/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":"2510.18784","created_at":"2026-06-19T16:12:48.672920+00:00"},{"alias_kind":"arxiv_version","alias_value":"2510.18784v3","created_at":"2026-06-19T16:12:48.672920+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2510.18784","created_at":"2026-06-19T16:12:48.672920+00:00"},{"alias_kind":"pith_short_12","alias_value":"U72GGMCAE7J3","created_at":"2026-06-19T16:12:48.672920+00:00"},{"alias_kind":"pith_short_16","alias_value":"U72GGMCAE7J3GMPF","created_at":"2026-06-19T16:12:48.672920+00:00"},{"alias_kind":"pith_short_8","alias_value":"U72GGMCA","created_at":"2026-06-19T16:12:48.672920+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2605.17471","citing_title":"WinQ: Accelerating Quantization-Aware Training of Language Models Around Saddle Points","ref_index":32,"is_internal_anchor":true},{"citing_arxiv_id":"2604.03420","citing_title":"Zero-Shot Quantization via Weight-Space Arithmetic","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2604.09220","citing_title":"TinyNeRV: Compact Neural Video Representations via Capacity Scaling, Distillation, and Low-Precision Inference","ref_index":67,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/U72GGMCAE7J3GMPF7W3AYPXJDU","json":"https://pith.science/pith/U72GGMCAE7J3GMPF7W3AYPXJDU.json","graph_json":"https://pith.science/api/pith-number/U72GGMCAE7J3GMPF7W3AYPXJDU/graph.json","events_json":"https://pith.science/api/pith-number/U72GGMCAE7J3GMPF7W3AYPXJDU/events.json","paper":"https://pith.science/paper/U72GGMCA"},"agent_actions":{"view_html":"https://pith.science/pith/U72GGMCAE7J3GMPF7W3AYPXJDU","download_json":"https://pith.science/pith/U72GGMCAE7J3GMPF7W3AYPXJDU.json","view_paper":"https://pith.science/paper/U72GGMCA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2510.18784&json=true","fetch_graph":"https://pith.science/api/pith-number/U72GGMCAE7J3GMPF7W3AYPXJDU/graph.json","fetch_events":"https://pith.science/api/pith-number/U72GGMCAE7J3GMPF7W3AYPXJDU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/U72GGMCAE7J3GMPF7W3AYPXJDU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/U72GGMCAE7J3GMPF7W3AYPXJDU/action/storage_attestation","attest_author":"https://pith.science/pith/U72GGMCAE7J3GMPF7W3AYPXJDU/action/author_attestation","sign_citation":"https://pith.science/pith/U72GGMCAE7J3GMPF7W3AYPXJDU/action/citation_signature","submit_replication":"https://pith.science/pith/U72GGMCAE7J3GMPF7W3AYPXJDU/action/replication_record"}},"created_at":"2026-06-19T16:12:48.672920+00:00","updated_at":"2026-06-19T16:12:48.672920+00:00"}