{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:U24DAAFGWPOF6YBMOMOMYKSSWA","short_pith_number":"pith:U24DAAFG","schema_version":"1.0","canonical_sha256":"a6b83000a6b3dc5f602c731ccc2a52b03f8008d61bc7a03f894cf90f15a180c5","source":{"kind":"arxiv","id":"2207.00112","version":1},"attestation_state":"computed","paper":{"title":"Language model compression with weighted low-rank factorization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Hongxia Jin, Qian Lou, Sungen Chang, Ting Hua, Yen-Chang Hsu, Yilin Shen","submitted_at":"2022-06-30T21:57:07Z","abstract_excerpt":"Factorizing a large matrix into small matrices is a popular strategy for model compression. Singular value decomposition (SVD) plays a vital role in this compression strategy, approximating a learned matrix with fewer parameters. However, SVD minimizes the squared error toward reconstructing the original matrix without gauging the importance of the parameters, potentially giving a larger reconstruction error for those who affect the task accuracy more. In other words, the optimization objective of SVD is not aligned with the trained model's task accuracy. We analyze this previously unexplored "},"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":"2207.00112","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2022-06-30T21:57:07Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"4d8d652d4fcfa64221d23c0b1b36ab7539754033eceb67d09add5c0ae5a73e7d","abstract_canon_sha256":"7530c6873eb496345fbe02fb3d53b91651d4ab01b7a2ee154b141757b723133d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:36:39.191131Z","signature_b64":"9kiafoQNoIkh/l/nxypfahDpnHOjNQRQ6pYiT+aGj77oFC39TsJvE4kRGOghnNqidQKvvWN36VslE8PhLzHIAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a6b83000a6b3dc5f602c731ccc2a52b03f8008d61bc7a03f894cf90f15a180c5","last_reissued_at":"2026-07-05T04:36:39.190676Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:36:39.190676Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Language model compression with weighted low-rank factorization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Hongxia Jin, Qian Lou, Sungen Chang, Ting Hua, Yen-Chang Hsu, Yilin Shen","submitted_at":"2022-06-30T21:57:07Z","abstract_excerpt":"Factorizing a large matrix into small matrices is a popular strategy for model compression. Singular value decomposition (SVD) plays a vital role in this compression strategy, approximating a learned matrix with fewer parameters. However, SVD minimizes the squared error toward reconstructing the original matrix without gauging the importance of the parameters, potentially giving a larger reconstruction error for those who affect the task accuracy more. In other words, the optimization objective of SVD is not aligned with the trained model's task accuracy. We analyze this previously unexplored "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2207.00112","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/2207.00112/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":"2207.00112","created_at":"2026-07-05T04:36:39.190758+00:00"},{"alias_kind":"arxiv_version","alias_value":"2207.00112v1","created_at":"2026-07-05T04:36:39.190758+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2207.00112","created_at":"2026-07-05T04:36:39.190758+00:00"},{"alias_kind":"pith_short_12","alias_value":"U24DAAFGWPOF","created_at":"2026-07-05T04:36:39.190758+00:00"},{"alias_kind":"pith_short_16","alias_value":"U24DAAFGWPOF6YBM","created_at":"2026-07-05T04:36:39.190758+00:00"},{"alias_kind":"pith_short_8","alias_value":"U24DAAFG","created_at":"2026-07-05T04:36:39.190758+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23568","citing_title":"SVD-Surgeon: Optimal Singular-Value Surgery for Large Language Model Compression","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2606.19993","citing_title":"Activation- and Influence-Aware Ranks (AIR): Function-Preserving SVD Compression for LLMs","ref_index":180,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07596","citing_title":"Shortcuts in the Tail: Debiasing via Post-Hoc Spectral Compression of Fine-Tuning Updates","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00573","citing_title":"LASER: Loss-Aware Singular-value Decomposition and Rank Allocation for Efficient Low-Precision Vision-Language Models","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00535","citing_title":"DREAM-S: Speculative Decoding with Searchable Drafting and Target-Aware Refinement for Multimodal Generation","ref_index":53,"is_internal_anchor":false},{"citing_arxiv_id":"2505.12942","citing_title":"A3 : an Analytical Low-Rank Approximation Framework for Attention","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16600","citing_title":"Where Pretraining writes and Alignment reads: the asymmetry of Transformer weight space","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17985","citing_title":"SAFE-SVD: Sensitivity-Aware Fidelity-Enforcing SVD for Physics Foundation Models","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19842","citing_title":"Fast Tensorization of Neural Networks via Slice-wise Feature Distillation","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2510.25977","citing_title":"NeuronMLP: Efficient LLM Inference via Singular Value Decomposition Compression and Tiling on AWS Trainium","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2604.02570","citing_title":"WSVD: Weighted Low-Rank Approximation for Fast and Efficient Execution of Low-Precision Vision-Language Models","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08314","citing_title":"FlashSVD v1.5: Making Low-Rank Transformers Inference Actually Fast","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08568","citing_title":"Different Prompts, Different Ranks: Prompt-aware Dynamic Rank Selection for SVD-based LLM Compression","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02905","citing_title":"eOptShrinkQ: Near-Lossless KV Cache Compression Through Optimal Spectral Denoising and Quantization","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02829","citing_title":"Compress Then Adapt? No, Do It Together via Task-aware Union of Subspaces","ref_index":16,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/U24DAAFGWPOF6YBMOMOMYKSSWA","json":"https://pith.science/pith/U24DAAFGWPOF6YBMOMOMYKSSWA.json","graph_json":"https://pith.science/api/pith-number/U24DAAFGWPOF6YBMOMOMYKSSWA/graph.json","events_json":"https://pith.science/api/pith-number/U24DAAFGWPOF6YBMOMOMYKSSWA/events.json","paper":"https://pith.science/paper/U24DAAFG"},"agent_actions":{"view_html":"https://pith.science/pith/U24DAAFGWPOF6YBMOMOMYKSSWA","download_json":"https://pith.science/pith/U24DAAFGWPOF6YBMOMOMYKSSWA.json","view_paper":"https://pith.science/paper/U24DAAFG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2207.00112&json=true","fetch_graph":"https://pith.science/api/pith-number/U24DAAFGWPOF6YBMOMOMYKSSWA/graph.json","fetch_events":"https://pith.science/api/pith-number/U24DAAFGWPOF6YBMOMOMYKSSWA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/U24DAAFGWPOF6YBMOMOMYKSSWA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/U24DAAFGWPOF6YBMOMOMYKSSWA/action/storage_attestation","attest_author":"https://pith.science/pith/U24DAAFGWPOF6YBMOMOMYKSSWA/action/author_attestation","sign_citation":"https://pith.science/pith/U24DAAFGWPOF6YBMOMOMYKSSWA/action/citation_signature","submit_replication":"https://pith.science/pith/U24DAAFGWPOF6YBMOMOMYKSSWA/action/replication_record"}},"created_at":"2026-07-05T04:36:39.190758+00:00","updated_at":"2026-07-05T04:36:39.190758+00:00"}