{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003537013,0.0002135844,0.0002467782,0.0001221314,0.0002707212,0.000158763,0.0006363917,0.00009178087,0.00004537758],"category_scores_gemma":[0.0001110168,0.000203164,0.0001161421,0.0003998092,0.00003643204,0.0002881821,0.0004124196,0.0002252757,0.0002028382],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001066059,"about_ca_system_score_gemma":0.0001124742,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003904154,"about_ca_topic_score_gemma":0.00004188738,"domain_scores_codex":[0.9983759,0.00003609075,0.0003293168,0.0005349987,0.0001850995,0.0005386259],"domain_scores_gemma":[0.9989353,0.0001839852,0.00009085717,0.0005552376,0.0001328503,0.0001017768],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001084018,0.000636828,0.004334735,0.0004433905,0.0002708998,0.00009708227,0.002137081,0.2060003,0.0008255495,0.2235929,0.1342708,0.427282],"study_design_scores_gemma":[0.001279237,0.0002148892,0.001653777,0.0001040554,0.00002040833,0.000004338591,0.00001138284,0.612074,0.0002607407,0.001215552,0.3828628,0.0002987609],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0007593156,0.001088968,0.9866596,0.003232641,0.0006409078,0.0005026888,0.00000148281,0.0004220474,0.00669236],"genre_scores_gemma":[0.8573434,0.0002925061,0.08196023,0.005079991,0.0001839284,0.0004045571,0.00002424668,0.00003596933,0.05467516],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9046994,"threshold_uncertainty_score":0.8284787,"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."}}