{"id":"W7092599726","doi":"10.1002/aaai.70021","title":"CAGE challenge 4: A scalable multi‐agent reinforcement learning gym for autonomous cyber defence","year":2025,"lang":"en","type":"article","venue":"AI Magazine","topic":"Software-Defined Networks and 5G","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Defence Research and Development Canada","funders":"","keywords":"Reinforcement learning; Scalability; Marl; Adversarial system; Autonomy; Usability; Enterprise private network","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":[0.001478586,0.0004913613,0.0005281234,0.0001973579,0.0004529335,0.0005544284,0.001375178,0.001190319,0.003481994],"category_scores_gemma":[0.002425692,0.0001916727,0.0004191222,0.0001141587,0.0008308791,0.0008428921,0.001464959,0.001889368,0.0005306299],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008060383,"about_ca_system_score_gemma":0.001251478,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003222052,"about_ca_topic_score_gemma":0.003792142,"domain_scores_codex":[0.9995382,0.0001782806,0.00001591445,0.00008507525,0.0001076455,0.00007486893],"domain_scores_gemma":[0.9985373,0.0007281516,0.00009496674,0.0001918577,0.0002141099,0.000233586],"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":[0.0004784746,0.0007119341,0.001966084,0.0001549677,0.00008837799,0.0002071496,0.00009646883,0.918695,0.008024543,0.01226149,0.01244552,0.04486999],"study_design_scores_gemma":[0.00007961589,0.0003113048,0.0002384084,0.000005159334,0.000005254599,0.0000155626,0.00002110726,0.9928433,0.001656496,0.002346177,0.002470569,0.000006975474],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5201547,0.0006465815,0.4270352,0.004807972,0.0007663984,0.00118272,0.0008302471,0.006519486,0.03805668],"genre_scores_gemma":[0.9067078,0.00006965639,0.08667523,0.0004372136,0.00003393622,0.0003213122,0.0003555501,0.000110773,0.005288486],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003481994,"threshold_uncertainty_score":0.01164842,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01904198579425885,"score_gpt":0.2668162894243775,"score_spread":0.2477743036301186,"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."}}