{"id":"W4394838778","doi":"10.1145/3618260.3649784","title":"On the Power of Interactive Proofs for Learning","year":2024,"lang":"en","type":"preprint","venue":"","topic":"Machine Learning and Algorithms","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Qualcomm (Canada); Simon Fraser University","funders":"Engineering and Physical Sciences Research Council; Natural Sciences and Engineering Research Council of Canada; University of Warwick; UK Research and Innovation","keywords":"Mathematical proof; Computer science; Power (physics); Mathematics","routes":{"ca_aff":true,"ca_fund":true,"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":[0.01446896,0.002312956,0.002217812,0.001998877,0.003586047,0.007598991,0.006381681,0.004464964,0.01312166],"category_scores_gemma":[0.07171541,0.001840267,0.003070391,0.001882402,0.01267601,0.02754724,0.01222939,0.01576179,0.002413587],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005204855,"about_ca_system_score_gemma":0.003346415,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001606767,"about_ca_topic_score_gemma":0.001074875,"domain_scores_codex":[0.9792669,0.008833121,0.0007615816,0.003320812,0.00610377,0.001713756],"domain_scores_gemma":[0.829737,0.1369129,0.003590033,0.02520153,0.002768646,0.001789894],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002770006,0.00009141656,0.0004214866,0.0002035606,0.00005438187,0.0001069661,0.0003738879,0.031932,0.001250604,0.9474632,0.001432616,0.01639283],"study_design_scores_gemma":[0.00008477005,0.00005288721,0.00008179287,0.00006376264,0.00002691936,0.00005728115,0.0000298709,0.1202276,0.00156904,0.8744912,0.00328221,0.00003280301],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02406814,0.001579606,0.9390118,0.005030382,0.000148453,0.0002010503,0.0002645752,0.001631652,0.0280644],"genre_scores_gemma":[0.7257785,0.001950971,0.2551821,0.001960176,0.0008931331,0.0009326874,0.0005071426,0.0009814888,0.01181376],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01446896,"threshold_uncertainty_score":0.07652014,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01367677723551281,"score_gpt":0.2914392437634967,"score_spread":0.2777624665279839,"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."}}