{"id":"W4401023825","doi":"10.24963/ijcai.2024/616","title":"StarFT: Robust Fine-tuning of Zero-shot Models via Spuriosity Alignment","year":2024,"lang":"en","type":"article","venue":"","topic":"VLSI and Analog Circuit Testing","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Kootenay Association for Science & Technology","funders":"National Science and Technology Major Project","keywords":"Adaptation (eye); Computer science; Test (biology); Transfer (computing); Parallel computing; Physics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001864941,0.001741419,0.001246672,0.0009460966,0.0006023926,0.001452577,0.003042114,0.001816886,0.002664996],"category_scores_gemma":[0.006778044,0.0005180617,0.001248029,0.0005951719,0.001365975,0.002522719,0.002581281,0.003406435,0.001663853],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001121554,"about_ca_system_score_gemma":0.001230062,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0046226,"about_ca_topic_score_gemma":0.006256802,"domain_scores_codex":[0.9989002,0.0002764302,0.00004151211,0.0004369933,0.0002067301,0.0001381564],"domain_scores_gemma":[0.9982069,0.0007897358,0.0001843151,0.0004230587,0.0002361303,0.000159936],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0005919387,0.0004393624,0.003871195,0.0003300218,0.0002683803,0.0002597877,0.0004975666,0.4869007,0.03344588,0.01722027,0.01368921,0.4424857],"study_design_scores_gemma":[0.00001599374,0.00006630619,0.0002137145,0.00001271558,0.00001270523,0.00004194353,0.00003371858,0.9858269,0.003985376,0.008805281,0.0009703989,0.00001502366],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.05109802,0.0005423471,0.9390194,0.0002972228,0.000126821,0.00008315346,0.0002672704,0.006387551,0.002178184],"genre_scores_gemma":[0.7379959,0.0003704792,0.2474164,0.0009350901,0.0002088129,0.0002606359,0.002897089,0.0014963,0.008419244],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.0046226,"threshold_uncertainty_score":0.00986284,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06075186670880472,"score_gpt":0.2475930412779976,"score_spread":0.1868411745691929,"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."}}