{"id":"W4412122873","doi":"10.1073/pnas.2502599122","title":"Asymptotic theory of in-context learning by linear attention","year":2025,"lang":"en","type":"article","venue":"Proceedings of the National Academy of Sciences","topic":"Domain Adaptation and Few-Shot Learning","field":"Computer Science","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada); Perimeter Institute","funders":"National Science Foundation","keywords":"Learning curve; Computer science; Generalization; Memorization; Context (archaeology); Security token; Task (project management); Scaling; Mathematics; Artificial intelligence; Machine learning; Cognitive psychology; Psychology","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":[0.002265264,0.0008363226,0.00109349,0.0009698693,0.0006228054,0.001140694,0.002261753,0.001490706,0.005025387],"category_scores_gemma":[0.02037521,0.000712934,0.0007725573,0.0004195739,0.002452776,0.003509488,0.001979515,0.002340637,0.0009802249],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001991416,"about_ca_system_score_gemma":0.001030123,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005032755,"about_ca_topic_score_gemma":0.00403199,"domain_scores_codex":[0.9991745,0.0002786785,0.00003544487,0.0001653879,0.0002044044,0.0001415224],"domain_scores_gemma":[0.9933612,0.004697573,0.0004984006,0.0006305302,0.0005415133,0.0002706991],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001283731,0.000123018,0.002811105,0.0003621654,0.00009256365,0.0002160942,0.0004537008,0.4129357,0.007713572,0.5344504,0.004441392,0.03627197],"study_design_scores_gemma":[0.00001107818,0.00003196657,0.0005235343,0.00001992833,0.000009508764,0.0000527996,0.00001868371,0.8617915,0.0005546387,0.1365421,0.0004305096,0.00001362963],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0978682,0.0009396581,0.8822783,0.002060266,0.0000782901,0.00007408371,0.0001731722,0.001229233,0.01529874],"genre_scores_gemma":[0.9265231,0.000816491,0.05811663,0.0006269105,0.0001408106,0.0003424505,0.0002418157,0.0003100036,0.01288171],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.005032755,"threshold_uncertainty_score":0.01681161,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02662637391215992,"score_gpt":0.2931535029031697,"score_spread":0.2665271289910098,"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."}}