{"id":"W4414561732","doi":"10.1029/2025jh000769","title":"Leveraging Sparse Autoencoders to Reveal Interpretable Features in Geophysical Models","year":2025,"lang":"en","type":"article","venue":"Journal of Geophysical Research Machine Learning and Computation","topic":"Reservoir Engineering and Simulation Methods","field":"Engineering","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; Jet Propulsion Laboratory; National Aeronautics and Space Administration","keywords":"Perceptron; Set (abstract data type); Feature (linguistics); Artificial neural network; Aggregate (composite); Pattern recognition (psychology); Representation (politics); ENCODE","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":[0.001190453,0.0001237115,0.0002914894,0.0006370863,0.00009047319,0.0001136556,0.0001339251,0.00005724929,0.000001910409],"category_scores_gemma":[0.000341021,0.0001138803,0.00006461995,0.0007082284,0.00002949034,0.0002098764,0.00005568882,0.001340567,0.000002889444],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001088214,"about_ca_system_score_gemma":0.00004114559,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000135378,"about_ca_topic_score_gemma":0.000004171168,"domain_scores_codex":[0.9985014,0.0003208248,0.0003277228,0.0001472124,0.0004082747,0.0002946057],"domain_scores_gemma":[0.9988503,0.0007282726,0.00003455495,0.00006776317,0.000187948,0.0001311249],"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":[0.00007372911,0.00002864808,0.0005945954,0.00007403947,0.00002596915,0.00001172398,0.0008204466,0.97128,0.001005472,0.0003317697,0.0002431288,0.02551048],"study_design_scores_gemma":[0.0004676704,0.0001374297,0.01423445,0.0002483656,0.000005567394,0.000004647492,0.0001121183,0.9747838,0.00009222099,0.009452906,0.0003702244,0.00009062455],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.632995,0.000266841,0.3654761,0.0003337096,0.0001001519,0.0000719368,3.749268e-7,0.00003666886,0.000719222],"genre_scores_gemma":[0.9851105,0.00005885493,0.01425739,0.00002037443,0.00009068847,0.000003884279,0.000002048475,0.00001619662,0.000440089],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3521155,"threshold_uncertainty_score":0.582417,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0310335110787255,"score_gpt":0.3672593779348382,"score_spread":0.3362258668561127,"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."}}