Design parameters of proposed NES machine learning method
| Parameter Description | Value |
|---|---|
| Gaussian proposal variance (σ2) | 0.1 |
| learning rate (for ℓ < 500) | 10−4 |
| learning rate (for ℓ > 500) | 5 · 10−5 |
| hidden layer size | 2K ℓ |
| number hidden layers | 2 |
| output activation | tanh |
| network optimizer | Adam (Kingma & Ba, 2015) |
| batch size (N) | 100 |
| number iterations | 10,000 |