{"id":"W4387324024","doi":"10.48550/arxiv.2310.00893","title":"Engineering the Neural Collapse Geometry of Supervised-Contrastive Loss","year":2023,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Visual Attention and Saliency Detection","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; King Abdullah University of Science and Technology; Universities Space Research Association; National Science Foundation","keywords":"Embedding; Limiting; Artificial intelligence; Computer science; Classifier (UML); Entropy (arrow of time); Benchmark (surveying); Feature (linguistics); Geometry; Cross entropy; Artificial neural network; Feature vector; Pattern recognition (psychology); Machine learning; Mathematics; Algorithm; Engineering; Mechanical engineering","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":[],"consensus_categories":[],"category_scores_codex":[0.0002715351,0.0001971654,0.0002327544,0.0002999306,0.0001109788,0.00007049116,0.001275385,0.0001497734,0.00001579811],"category_scores_gemma":[0.00006568845,0.0001829103,0.0002201995,0.001170245,0.00007665516,0.0002207179,0.001134467,0.0004135666,0.00006361875],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008544594,"about_ca_system_score_gemma":0.0000646677,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001248313,"about_ca_topic_score_gemma":0.00002179803,"domain_scores_codex":[0.9987757,0.00008889556,0.0001930819,0.0005827419,0.0001158548,0.0002436773],"domain_scores_gemma":[0.9987743,0.0001547917,0.0001555666,0.0006750931,0.0001604042,0.00007978154],"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":[0.0000370381,0.0001051184,0.00594114,0.0001710061,0.0001895313,0.0002603398,0.0005876787,0.9095089,0.0004984471,0.08147673,0.0002205748,0.001003524],"study_design_scores_gemma":[0.0002639329,0.0000476278,0.01677022,0.00004906527,0.00003569375,0.000005079205,0.0001115485,0.9796554,0.0003928966,0.002388257,0.00006322866,0.0002170028],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6304854,0.00001774643,0.367741,0.000128733,0.001120221,0.0001812954,0.00001230648,0.0002186603,0.00009463255],"genre_scores_gemma":[0.9988379,0.00004774703,0.0001293595,0.00003311738,0.00005206126,0.000001192646,0.000004680277,0.00001350702,0.0008804334],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3683525,"threshold_uncertainty_score":0.7458865,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06787373052823772,"score_gpt":0.1952296738075231,"score_spread":0.1273559432792854,"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."}}