Cyber Defense on NetForge_RL MockHypervisor last 20% (train convergence)
57,135Blue RewardCT-GMARL
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| CT-GMARLArchitecture=CT-GMARL (Ours)2026.04 | 57,135 | 2,475 | 144 | 1 | 4 | 0.0031 | 10 | |
| CT-GMARL (No-Beta)Ablation=No-Beta2026.04 | 56,656 | 46 | 48 | 0 | 4 | 0.0055 | 7.4 | |
| CT-GMARL (No-GAT)Ablation=No-GAT2026.04 | 56,565 | 38 | 73 | 0 | 4 | 0.0011 | 4.8 | |
| CT-GMARL (No-ODE)Ablation=No-ODE2026.04 | 51,783 | 46 | 32 | 0 | — | 0.0026 | 5 | |
| R-MAPPOArchitecture=R-MAPPO Baseline2026.04 | 28,347 | 2 | 13 | 0 | — | 0.0067 | 17.2 | |
| QMIXArchitecture=QMIX Baseline2026.04 | 26,649 | 0 | 5 | 0 | — | 0.0627 | 11.7 |