{"id":"W3132401450","doi":"10.1162/neco_a_01557","title":"How to Represent Part-Whole Hierarchies in a Neural Network","year":2022,"lang":"en","type":"article","venue":"Neural Computation","topic":"Neural Networks and Applications","field":"Computer Science","cited_by":149,"is_retracted":false,"has_abstract":true,"ca_institutions":"Vector Institute; University of Toronto","funders":"","keywords":"Interpretability; Parsing; Transformer; Computer science; Artificial intelligence; Artificial neural network; Hierarchy; Representation (politics); Parse tree; Natural language processing; Theoretical computer science; Machine learning; Engineering; Voltage","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":[],"consensus_categories":[],"category_scores_codex":[0.0001768463,0.0001304472,0.0001457003,0.0001177299,0.0003891449,0.0002805399,0.0005943299,0.00001730499,0.000005697581],"category_scores_gemma":[0.00001313653,0.0001354857,0.00005711043,0.001371925,0.00001877132,0.0003218913,0.000651821,0.0002791756,0.000008350656],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005483865,"about_ca_system_score_gemma":0.00001765309,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002610048,"about_ca_topic_score_gemma":0.00002773886,"domain_scores_codex":[0.9983702,0.0002154349,0.0002244626,0.0004969927,0.0003308163,0.000362042],"domain_scores_gemma":[0.9992976,0.0001429246,0.00008524137,0.0003407005,0.00003262189,0.000100942],"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":[0.000008518702,0.00003438669,0.0005682677,0.000002219221,0.000001728474,0.00001814045,0.0002231913,0.8738377,0.0001344868,0.002940248,0.01414036,0.1080907],"study_design_scores_gemma":[0.0001943477,0.0001385073,0.00847872,0.000004035924,0.000001672397,0.0000233306,0.00003540527,0.9527158,0.00002099342,0.005485818,0.03273733,0.0001640727],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6751882,0.0002829992,0.2387245,0.08226894,0.001546535,0.001144933,0.00001047745,0.0004528078,0.0003805761],"genre_scores_gemma":[0.9919261,0.000001745445,0.005146534,0.002058707,0.0002629569,0.0002446381,0.00002389324,0.00001097019,0.0003245028],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3167378,"threshold_uncertainty_score":0.5524946,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02746012427718318,"score_gpt":0.2652115562248038,"score_spread":0.2377514319476206,"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."}}