{"id":"W4401281176","doi":"10.1016/j.csl.2024.101695","title":"Speech self-supervised representations benchmarking: A case for larger probing heads","year":2024,"lang":"en","type":"article","venue":"Computer Speech & Language","topic":"Speech Recognition and Synthesis","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":false,"ca_institutions":"Concordia University; Mila - Quebec Artificial Intelligence Institute","funders":"Agence de l'innovation de Défense","keywords":"Benchmarking; Computer science; Ranking (information retrieval); Inference; Task (project management); Downstream (manufacturing); Generalization; Feature (linguistics); Set (abstract data type); Artificial intelligence; Architecture; Machine learning; Benchmark (surveying); Data set; Natural language processing","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":["metaepi_narrow","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.0006361717,0.0003151436,0.0003158851,0.000400098,0.0002849005,0.001165055,0.0007437792,0.0001324046,0.0003113483],"category_scores_gemma":[0.00004443881,0.0002925408,0.0003190278,0.0008188226,0.00003278946,0.0007019768,0.0003408784,0.0002516861,0.0002381973],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001023507,"about_ca_system_score_gemma":0.0001329153,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006665369,"about_ca_topic_score_gemma":0.0000501552,"domain_scores_codex":[0.9975308,0.0001280109,0.0004464518,0.0009713273,0.0003514787,0.0005719031],"domain_scores_gemma":[0.9982301,0.0004611732,0.00006875436,0.0008783738,0.0001493296,0.0002122978],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000004818143,0.0001549872,0.00002955548,0.0001773877,0.0001278164,0.01098001,0.006668122,0.000008635706,0.001159789,0.00409542,0.01239457,0.9641989],"study_design_scores_gemma":[0.0007283291,0.0001313769,0.00008750348,0.0001798701,0.0000764618,0.01066945,0.0003567779,0.9455146,0.01484004,0.001080798,0.02561928,0.0007154884],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.05454756,0.0007905558,0.9337468,0.001898477,0.002306395,0.001095933,0.00004756067,0.002075882,0.003490799],"genre_scores_gemma":[0.08816654,0.00001663682,0.9087351,0.0008780144,0.001345278,0.0001501124,0.00004536579,0.00005386992,0.000609111],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9634834,"threshold_uncertainty_score":0.9999527,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02313783000429004,"score_gpt":0.2944255563429296,"score_spread":0.2712877263386396,"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."}}