Table 2 Accuracy of CABLE with different RL methods for task assignment
From: Building adaptive knowledge bases for evolving continual learning models
Methods | FMNIST | CIFAR100 | M-IN | CUB | CORe50 |
|---|---|---|---|---|---|
CABLE+Random | 17.36% | 18.55% | 15.23% | 18.11% | 17.22% |
CABLE+TRPO | 19.15% | 19.52% | 16.98% | 17.79% | 17.07% |
CABLE+PPO | 20.01% | 20.13% | 17.09% | 17.85% | 17.59% |
CABLE+A2C | 20.01% | 20.13% | 17.09% | 18.86% | 18.15% |
CABLE+A3C | 20.12% | 20.18% | 17.41% | 18.84% | 18.11% |
CABLE+Clipped PPO | 21.24% | 21.17% | 18.22% | 19.05% | 18.95% |
