{"id":"W4415332398","doi":"10.48550/arxiv.2509.23325","title":"Robust Fine-Tuning from Non-Robust Pretrained Models: Mitigating Suboptimal Transfer With Epsilon-Scheduling","year":2025,"lang":"en","type":"preprint","venue":"ArXiv.org","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; Consortium de Recherche et d’innovation en Aérospatiale au Québec","keywords":"Robustness (evolution); Transfer of learning; Metric (unit); Workflow; Task (project management); Schedule","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.0009456086,0.001225526,0.001297477,0.0004436928,0.0006948028,0.000576304,0.003827358,0.0008641419,0.00006779361],"category_scores_gemma":[0.0002918785,0.001222079,0.0004172063,0.0009164629,0.0002296572,0.001247973,0.002716872,0.004248196,0.00002677453],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003011456,"about_ca_system_score_gemma":0.0009580437,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001023248,"about_ca_topic_score_gemma":0.0001595497,"domain_scores_codex":[0.9935698,0.0003642301,0.001144251,0.00273575,0.0009637817,0.001222179],"domain_scores_gemma":[0.9957096,0.0007781904,0.0004510033,0.002320694,0.0004071993,0.000333303],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00006445292,0.00006603175,0.02511575,0.0001994501,0.0002995992,0.0001256131,0.003591079,0.9675646,0.0002714961,0.0009985322,0.00002379514,0.001679606],"study_design_scores_gemma":[0.001323291,0.00008556696,0.00399461,0.002193815,0.0001923836,0.000009150713,0.0002385559,0.9890918,0.0006806579,0.0008738073,0.00001715286,0.001299206],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3338833,0.0001881306,0.6608698,0.000651638,0.0009545729,0.0005646304,0.00003793331,0.000644865,0.002205112],"genre_scores_gemma":[0.5428219,0.00001574812,0.4561028,0.0001969344,0.0003463171,0.00008741185,0.0001154353,0.00007772903,0.0002356663],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.2089386,"threshold_uncertainty_score":0.9990229,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06255534979700363,"score_gpt":0.2593754850901944,"score_spread":0.1968201352931908,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}