{"id":"W4407200740","doi":"10.1007/s10994-024-06659-z","title":"Robust query performance prediction for dense retrievers via adaptive disturbance generation","year":2025,"lang":"en","type":"article","venue":"Machine Learning","topic":"Advanced Image and Video Retrieval Techniques","field":"Computer Science","cited_by":6,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Waterloo; Toronto Metropolitan University","funders":"","keywords":"Disturbance (geology); Computer science; Artificial intelligence; Control theory (sociology); Pattern recognition (psychology); Biology; Control (management)","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":[0.0009180484,0.0007630389,0.001121322,0.0006553971,0.0002789292,0.0009134402,0.001026236,0.0007275846,0.00100053],"category_scores_gemma":[0.005220745,0.0003230152,0.0003113772,0.0008145785,0.0005573669,0.001016944,0.0008403487,0.0009029244,0.0006101502],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007195512,"about_ca_system_score_gemma":0.0008273752,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005974723,"about_ca_topic_score_gemma":0.006488961,"domain_scores_codex":[0.9991834,0.0001097258,0.00003945676,0.0001748486,0.0003438135,0.0001489196],"domain_scores_gemma":[0.9979749,0.001013757,0.0002240079,0.0002807971,0.0004330393,0.00007347749],"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":[0.0007845068,0.0002210327,0.00355425,0.0001382578,0.00006450201,0.0001648003,0.00007684682,0.7271619,0.0378427,0.004010856,0.003386599,0.2225937],"study_design_scores_gemma":[0.000005342409,0.00002448948,0.0003612596,0.000001427569,0.000004405679,0.00001342403,0.000005021477,0.9967758,0.00215086,0.0005460468,0.0001078702,0.00000418237],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.06504699,0.0004708773,0.9316648,0.0002018483,0.00005662689,0.00006061682,0.0001734147,0.001215623,0.001109335],"genre_scores_gemma":[0.9489013,0.0001844529,0.04851734,0.00007296169,0.0000758586,0.00005142032,0.0004405738,0.00008841947,0.001667662],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.005974723,"threshold_uncertainty_score":0.01187986,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02459554223764634,"score_gpt":0.2541639193749147,"score_spread":0.2295683771372684,"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."}}