{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:Z7LJINGGZCQBOM2RNXH6T7WZZA","short_pith_number":"pith:Z7LJINGG","schema_version":"1.0","canonical_sha256":"cfd69434c6c8a01733516dcfe9fed9c8391726d9f5d40d04553e32ab0be99f63","source":{"kind":"arxiv","id":"2503.15748","version":1},"attestation_state":"computed","paper":{"title":"PARQ: Piecewise-Affine Regularized Quantization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.OC"],"primary_cat":"cs.LG","authors_text":"Aaron Defazio, Andrey Gromov, Jianhao Ma, Lin Xiao, Lisa Jin, Zechun Liu","submitted_at":"2025-03-19T23:38:49Z","abstract_excerpt":"We develop a principled method for quantization-aware training (QAT) of large-scale machine learning models. Specifically, we show that convex, piecewise-affine regularization (PAR) can effectively induce the model parameters to cluster towards discrete values. We minimize PAR-regularized loss functions using an aggregate proximal stochastic gradient method (AProx) and prove that it has last-iterate convergence. Our approach provides an interpretation of the straight-through estimator (STE), a widely used heuristic for QAT, as the asymptotic form of PARQ. We conduct experiments to demonstrate "},"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":"2503.15748","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-03-19T23:38:49Z","cross_cats_sorted":["math.OC"],"title_canon_sha256":"2d51669a835ac6d4e7070c222d6a61a8834f0d4fa39b298dc8b2e849eb5aa9d1","abstract_canon_sha256":"95a1dff4d3d8037d431596f07f79457683be484ec655b9d7056f064321ab48bd"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:35:57.757890Z","signature_b64":"l69U6zEAJ1qn7OR99OWWIzOcXP8w8GbtuYNHQlgqu63l+FKtNbAiAGrFf1f6YUAmrEMDDB/nnFGLvsWAjne3Cg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cfd69434c6c8a01733516dcfe9fed9c8391726d9f5d40d04553e32ab0be99f63","last_reissued_at":"2026-07-05T10:35:57.756820Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:35:57.756820Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"PARQ: Piecewise-Affine Regularized Quantization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.OC"],"primary_cat":"cs.LG","authors_text":"Aaron Defazio, Andrey Gromov, Jianhao Ma, Lin Xiao, Lisa Jin, Zechun Liu","submitted_at":"2025-03-19T23:38:49Z","abstract_excerpt":"We develop a principled method for quantization-aware training (QAT) of large-scale machine learning models. Specifically, we show that convex, piecewise-affine regularization (PAR) can effectively induce the model parameters to cluster towards discrete values. We minimize PAR-regularized loss functions using an aggregate proximal stochastic gradient method (AProx) and prove that it has last-iterate convergence. Our approach provides an interpretation of the straight-through estimator (STE), a widely used heuristic for QAT, as the asymptotic form of PARQ. We conduct experiments to demonstrate "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.15748","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/2503.15748/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":"2503.15748","created_at":"2026-07-05T10:35:57.756966+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.15748v1","created_at":"2026-07-05T10:35:57.756966+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.15748","created_at":"2026-07-05T10:35:57.756966+00:00"},{"alias_kind":"pith_short_12","alias_value":"Z7LJINGGZCQB","created_at":"2026-07-05T10:35:57.756966+00:00"},{"alias_kind":"pith_short_16","alias_value":"Z7LJINGGZCQBOM2R","created_at":"2026-07-05T10:35:57.756966+00:00"},{"alias_kind":"pith_short_8","alias_value":"Z7LJINGG","created_at":"2026-07-05T10:35:57.756966+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.10989","citing_title":"SURGE: Surrogate Gradient Adaptation in Binary Neural Networks","ref_index":129,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10989","citing_title":"SURGE: Surrogate Gradient Adaptation in Binary Neural Networks","ref_index":129,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Z7LJINGGZCQBOM2RNXH6T7WZZA","json":"https://pith.science/pith/Z7LJINGGZCQBOM2RNXH6T7WZZA.json","graph_json":"https://pith.science/api/pith-number/Z7LJINGGZCQBOM2RNXH6T7WZZA/graph.json","events_json":"https://pith.science/api/pith-number/Z7LJINGGZCQBOM2RNXH6T7WZZA/events.json","paper":"https://pith.science/paper/Z7LJINGG"},"agent_actions":{"view_html":"https://pith.science/pith/Z7LJINGGZCQBOM2RNXH6T7WZZA","download_json":"https://pith.science/pith/Z7LJINGGZCQBOM2RNXH6T7WZZA.json","view_paper":"https://pith.science/paper/Z7LJINGG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.15748&json=true","fetch_graph":"https://pith.science/api/pith-number/Z7LJINGGZCQBOM2RNXH6T7WZZA/graph.json","fetch_events":"https://pith.science/api/pith-number/Z7LJINGGZCQBOM2RNXH6T7WZZA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Z7LJINGGZCQBOM2RNXH6T7WZZA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Z7LJINGGZCQBOM2RNXH6T7WZZA/action/storage_attestation","attest_author":"https://pith.science/pith/Z7LJINGGZCQBOM2RNXH6T7WZZA/action/author_attestation","sign_citation":"https://pith.science/pith/Z7LJINGGZCQBOM2RNXH6T7WZZA/action/citation_signature","submit_replication":"https://pith.science/pith/Z7LJINGGZCQBOM2RNXH6T7WZZA/action/replication_record"}},"created_at":"2026-07-05T10:35:57.756966+00:00","updated_at":"2026-07-05T10:35:57.756966+00:00"}