Table 1.
Parameters from single-learning strategies are also present in dual-learning strategies excepts for θ, which disappears in entropy-based coordination and α which disappears in VPI-based selection.
| Model | Symbol | Range | Description |
|---|---|---|---|
| Q-L only | α | 0 < α < 1 | Learning rate |
| β | 0 < β < 100 | Softmax temperature | |
| BWM only | N | 1 < N < 10 | Memory size |
| θ | 0 < θ < log|A| | Fixed entropy threshold | |
| ϵ | 0 < ϵ < 0.1 | Memory items decay | |
| VPI-based selection | η | 0.00001 < η < 0.001 | Covariance initialization |
| σr | 0 < σr < 1 | Reward rate update | |
| Weight-based mixture | w0 | 0 < w0 < 1 | Initial weight |
| Entropy-based coordination | λ1, λ2 | 0.00001 < λi < 1000 | Sigmoide parameters |
| βfinal | 0 < βfinal < 100 | Softmax temperature | |
| σ | 0 < σ < 20 | Simulated RT |