{"id":"W2911290743","doi":"10.1073/pnas.1907377117","title":"On instabilities of deep learning in image reconstruction and the potential costs of AI","year":2020,"lang":"en","type":"article","venue":"Proceedings of the National Academy of Sciences","topic":"Medical Imaging Techniques and Applications","field":"Medicine","cited_by":759,"is_retracted":false,"has_abstract":true,"ca_institutions":"Simon Fraser University","funders":"Fundação para a Ciência e a Tecnologia; Natural Sciences and Engineering Research Council of Canada; European Commission; Research Councils UK; Royal Society; Engineering and Physical Sciences Research Council; Canadian Network for Research and Innovation in Machining Technology, Natural Sciences and Engineering Research Council of Canada; Leverhulme Trust; Nvidia","keywords":"Counterintuitive; Deep learning; Artificial intelligence; Computer science; Image (mathematics); Instability; Stability (learning theory); Iterative reconstruction; Field (mathematics); Sampling (signal processing); Computer vision; Machine learning; Mathematics; Physics","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007719711,0.00003564607,0.0001288336,0.0000617205,0.00004380745,0.000003832011,0.0001515237,0.00003012788,0.000009869469],"category_scores_gemma":[0.001391319,0.00001966685,0.00003240195,0.0003320654,0.00176097,0.0000990153,0.00005139571,0.000194218,6.770733e-8],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001020971,"about_ca_system_score_gemma":0.00001874255,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001085036,"about_ca_topic_score_gemma":2.984587e-8,"domain_scores_codex":[0.9991655,0.000006898839,0.0002439291,0.0000948633,0.0004420856,0.00004670254],"domain_scores_gemma":[0.9994707,0.0001220491,0.0002255893,0.000004890862,0.0001574266,0.00001931893],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0002054045,0.0000666889,0.01447557,0.0003131197,0.0000133635,6.248149e-9,0.001001538,0.0000566961,0.6629937,0.3114793,0.0002510281,0.009143499],"study_design_scores_gemma":[0.001809091,0.0003259761,0.06844246,0.0007950141,0.00005070885,0.0000296536,0.002873892,0.09080582,0.649829,0.1848678,0.00007601052,0.00009458411],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9577798,0.00003678025,0.00002175819,0.04004209,0.000003178834,0.0001797217,0.000002087565,0.000005966283,0.00192865],"genre_scores_gemma":[0.9963892,0.00004364084,0.003150912,0.0003841459,0.00001726162,0.000006284076,6.401196e-8,0.000001351793,0.000007140458],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1266115,"threshold_uncertainty_score":0.6488369,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02349296957277632,"score_gpt":0.3122047471846728,"score_spread":0.2887117776118965,"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."}}