{"id":"W4389520394","doi":"10.18653/v1/2023.emnlp-main.652","title":"Balance Act: Mitigating Hubness in Cross-Modal Retrieval with Query and Gallery Banks","year":2023,"lang":"en","type":"article","venue":"","topic":"Multimodal Machine Learning Applications","field":"Computer Science","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Computer science; Softmax function; Inference; Boosting (machine learning); Information retrieval; Normalization (sociology); Similarity (geometry); Dual (grammatical number); Data mining; Artificial intelligence; Image (mathematics); Deep learning","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004081921,0.0001252918,0.0001419928,0.0001260265,0.0001432383,0.0002409913,0.0004459643,0.00006098645,0.00001225319],"category_scores_gemma":[0.0001266531,0.0001070007,0.00001773242,0.001035487,0.00008408966,0.0003774687,0.0002709097,0.0002827655,0.000063621],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002381292,"about_ca_system_score_gemma":0.00006440512,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004165607,"about_ca_topic_score_gemma":0.00007451014,"domain_scores_codex":[0.998748,0.00004656627,0.0001889795,0.0004734813,0.0002357942,0.0003071952],"domain_scores_gemma":[0.9990733,0.0003087903,0.00006291473,0.0004220372,0.00004971922,0.00008320734],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001821895,0.00003364015,0.9686395,0.0000389306,0.000008955018,0.00003328894,0.00102376,0.004531382,0.002931994,0.01334388,0.00009796361,0.009298484],"study_design_scores_gemma":[0.0002792631,0.00001676905,0.6883475,0.00002330215,7.380515e-7,0.00001020019,0.00001839634,0.3102477,0.0003450034,0.0005396167,0.00005421483,0.000117256],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9733485,0.00001396488,0.02120798,0.001880588,0.00004319255,0.0001605108,0.000001241512,0.0004728999,0.002871145],"genre_scores_gemma":[0.9745099,0.000004605803,0.02471931,0.0001762475,0.00003377431,0.00001576481,0.000004022282,0.0000126285,0.0005237776],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3057164,"threshold_uncertainty_score":0.436336,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01178640720667072,"score_gpt":0.2904348914203238,"score_spread":0.2786484842136531,"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."}}