{"paper":{"title":"Wanda-Guided Sparse Perturbations for Efficient Zeroth-Order Optimization","arxiv_id":"2611.16317","primary_cat":"cs.LG","submitted_at":"2026-11-12T05:25:58+00:00","version_compiled":"v1"},"notice":"Every quoted sentence is the authors' text from the arXiv listing abstract. Labels and probabilities are the outputs of classifier-1.13.0 on the question bank abstract-p1-20260923, exactly as returned. No sentence in this record was written by a language model.","schema":"pith.compiled-record/v0","status":"compiled","abstract":{"source":"arXiv listing abstract v1","sentences":[{"i":0,"p":0.99,"id":"s0","own":false,"role":"background","text":"Zeroth-order (ZO) optimization enables memory-efficient fine-tuning of large language models through forward-only evaluations, but high-dimensional random perturbations can limit optimization efficiency.","hedge":{"p":0.89,"label":"asserted"},"own_p":0.06,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.09,"role_probabilities":{"gap":0.01,"method":0.0,"result":0.0,"background":0.99,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":1,"p":0.97,"id":"s1","own":false,"role":"background","text":"Sparse perturbations reduce the search dimension, making the selection of informative coordinates critical.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.07,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.03,"method":0.0,"result":0.0,"background":0.97,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":2,"p":0.51,"id":"s2","own":true,"role":"contribution","text":"We propose Wanda-Importance Sparsity-MeZO (WIS-MeZO), which uses Wanda weight--activation saliency to construct task-conditioned sparse perturbation masks while keeping the pretrained model dense.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.94,"main_claim_p":0.14,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.49,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.51,"interpretation":0.0}},{"i":3,"p":0.73,"id":"s3","own":true,"role":"contribution","text":"We analyze support quality through gradient energy capture and relate it to gradient estimation and convergence.","hedge":{"p":0.92,"label":"asserted"},"own_p":0.86,"main_claim_p":0.04,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.23,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.73,"interpretation":0.04}},{"i":4,"p":1.0,"id":"s4","own":true,"role":"result","text":"Across 16 benchmarks with LLaMA and Qwen models ranging from 3B to 32B parameters, WIS-MeZO achieves higher aggregate scores than MeZO and reaches comparable validation losses in fewer optimization steps.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.96,"main_claim_p":0.76,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":5,"p":0.66,"id":"s5","own":true,"role":"result","text":"Comparisons with random and magnitude-based sparse perturbations further highlight the importance of informative support selection.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.86,"main_claim_p":0.1,"quantitative":true,"quantitative_p":0.65,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.66,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.34}}]},"compiled":{"main_claim":{"p":1.0,"text":"Across 16 benchmarks with LLaMA and Qwen models ranging from 3B to 32B parameters, WIS-MeZO achieves higher aggregate scores than MeZO and reaches comparable validation losses in fewer optimization steps.","location":"abstract s4","hedge_strength":{"p":1.0,"label":"asserted"}},"provenance":{"route":{"via":"TypeSafe API","model":"classifier-1.13.0","endpoint":"https://api.typesafe.ai/v1/systemone","provider":"TypeSafe"},"packing":"one request per paper","requests":1,"input_tokens":4927,"cost_usd_reported":0.0},"compiled_at":"2026-11-13","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.04,"states_limitation":0.12,"human_participants":0.02,"controlled_comparison":0.94,"randomized_assignment":0.02,"states_quantitative_result":0.82,"compares_with_prior_results":0.98,"mentions_code_or_data_availability":0.02},"evidential_basis":{"p":0.86,"value":"observation or experiment","probabilities":{"argument":0.0,"literature":0.0,"proof_or_derivation":0.0,"analysis_of_existing_data":0.0,"observation_or_experiment":0.86,"simulation_or_computation":0.14}},"contribution_type":{"p":1.0,"value":"method or tool","probabilities":{"review":0.0,"method_or_tool":1.0,"empirical_study":0.0,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.0,"position_or_commentary":0.0}},"contribution_statement":{"p":1.0,"text":"We propose Wanda-Importance Sparsity-MeZO (WIS-MeZO), which uses Wanda weight--activation saliency to construct task-conditioned sparse perturbation masks while keeping the pretrained model dense.","location":"abstract s2"}},"coverage":{"tier":"abstract","levels_run":["P abstract"],"not_examined":["full text","figures and tables","references and citing sentences","relations to other papers"],"sources_read":["arXiv listing abstract v1"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":6,"sentences_in_abstract":6},"measured_accuracy":{"url":"https://pith.science/accuracy","measured_on":"2026-10-02","method":"two independent model families labelled a fresh random sample blind; the figure is how often the published label matches their consensus","sentence_role_agreement":0.92,"main_claim_agreement":0.908,"typed_citation_agreement":0.933,"typed_citation_agreement_95_interval":[0.917,0.945],"review_comments_accurate":0.758,"review_comments_accurate_95_interval":[0.684,0.82],"review_caught_admitted_error_found_or_related":0.448,"review_comments_judged":{"judged_by":["gpt-5.6-luna","grok-4.7"],"judged_against":"Pith's own text copy of each paper: text extracted from the PDF, the first 90,000 characters","measured_on":"2026-10-04","accurate":113,"decided":149,"sampled_reviews":200,"split_by_lane":false,"by_lane":"https://pith.science/accuracy#by-lane"}}}