{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009320861,0.0005874092,0.0005135684,0.0006602014,0.0007821617,0.002343969,0.001362883,0.001352626,0.00825177],"category_scores_gemma":[0.003955513,0.0005797106,0.0008849499,0.0007069733,0.001378707,0.007671921,0.00152207,0.002124759,0.002473621],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008899422,"about_ca_system_score_gemma":0.0009777597,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00503664,"about_ca_topic_score_gemma":0.006321492,"domain_scores_codex":[0.9995655,0.0001199696,0.00003501644,0.0001319689,0.0001053889,0.00004208438],"domain_scores_gemma":[0.9994823,0.0001869805,0.0000305029,0.0001684785,0.00009483703,0.00003695084],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001264324,0.00004626166,0.0007864309,0.0003005529,0.00008563804,0.0001405085,0.0006338932,0.1119181,0.01116282,0.5197828,0.01316923,0.3418473],"study_design_scores_gemma":[0.0000214892,0.00002792959,0.0001651596,0.00007734219,0.00003416862,0.00009547144,0.0001179898,0.4085647,0.005563216,0.5657915,0.01950874,0.00003223109],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.007412868,0.0003747926,0.981922,0.001541722,0.0001148866,0.00005671028,0.0002422426,0.001708032,0.006626842],"genre_scores_gemma":[0.1831658,0.0008144861,0.8068721,0.0005682099,0.00007164959,0.0001798049,0.0004944156,0.0004995433,0.007333979],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.00825177,"threshold_uncertainty_score":0.02760488,"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."}}