Multiple model-based reinforcement learning
Neural Computation
Adaptive mixtures of local experts
Neural Computation
| Hi-index | 0.00 |
Robustness of a reinforcement-learning controller can be increased if multiple copies of an actor-critic controller can be made on-line, stored, and retrieved, thus preventing catastrophic forgetting in a non-stationary and non-linear environment. This confirms a previous proposal that neuronal replication evolved as a mechanism for memory.