{"id":"W4403407815","doi":"10.1109/icc51166.2024.10622985","title":"On the Out-of-Distribution Evaluation of ML-Based End-to-End Communications Systems","year":2024,"lang":"en","type":"article","venue":"","topic":"Error Correcting Code Techniques","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Manitoba","funders":"Futurewei Technologies","keywords":"End-to-end principle; Computer science; End user; Telecommunications; Computer network; World Wide Web","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":[0.006615988,0.001121828,0.0007809232,0.001087337,0.0004776068,0.001439109,0.0007972877,0.001279135,0.001150413],"category_scores_gemma":[0.0313019,0.0002079965,0.000273669,0.0006451584,0.001337233,0.002391744,0.002063569,0.0009768797,0.0003978441],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001332174,"about_ca_system_score_gemma":0.0007892664,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001781867,"about_ca_topic_score_gemma":0.001304189,"domain_scores_codex":[0.994818,0.002503081,0.0002781683,0.0004084042,0.001688613,0.0003036856],"domain_scores_gemma":[0.9754869,0.01840727,0.001379369,0.001489279,0.002950596,0.0002866114],"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":[0.001432702,0.0002182119,0.008693339,0.0003288867,0.0001361691,0.0002333918,0.0001508156,0.8315235,0.01726531,0.008236719,0.001001489,0.1307794],"study_design_scores_gemma":[0.00001807711,0.0004345832,0.002453938,0.00003576747,0.0000265764,0.0002564442,0.00006086299,0.9632263,0.03040927,0.002538266,0.0005141995,0.00002576423],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3079574,0.001696286,0.6805794,0.0007399506,0.0001223458,0.0001102835,0.0002113358,0.001320471,0.007262511],"genre_scores_gemma":[0.9654213,0.0003847495,0.03279383,0.0001679418,0.00003259516,0.00003036917,0.0001724407,0.00009479533,0.0009019988],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.006615988,"threshold_uncertainty_score":0.03498906,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1124957316790551,"score_gpt":0.3680758551219959,"score_spread":0.2555801234429408,"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."}}