Table A1.
Training parameters of the goal-approaching controller.
| Goal-approaching | Number of training | 100 |
| Episodes | ||
| Number of test labeled | 500 | |
| Data per episode | ||
| Syn1 initial weights | ||
| Inhibitory (20%) | −4 ± 1 | |
| Excitatory (80%) | 21.5 ± 3.5 | |
| Syn2 initial weights | ||
| Inhibitory (20%) | −2.5 ± 0.5 | |
| Excitatory (80%) | 4 ± 1 | |
| Max. learning rate | ηmax = 0.2 | |
| Min. learning rate | ηmin = 0.05 | |
| Amplitude of weight change | A+ = 0.4 | |
| for facilitation | ||
| Time window size for | τ+ = 10ms | |
| faciliation | ||
| Amplitude of weight change | A− = 1.05A+ = 0.42 | |
| for depression | ||
| Time window size for | τ+ = 10ms | |
| depression | ||
| Maximum weight | wmax = 50 | |
| Minimum weight | −wmax/2 = −25 | |
| Eligibility trace | c1 = 1/wmax = 0.02 | |
| constants | c2 = 1 | |
| Maximum output | ymax = 5 | |
| Target-facing neuron | Initial weights | 15 |
| Max. learning rate | ηmax = 0.2 | |
| Min. learning rate | ηmin = 0.05 | |
| Amplitude of weight change | A+ = 0.1 | |
| for facilitation | ||
| Time window size for | τ+ = 10ms | |
| faciliation | ||
| Amplitude of weight change | A− = 1.05A+ = 0.105 | |
| for depression | ||
| Time window size for | τ+ = 10ms | |
| depression | ||
| Maximum weight | wmax = 50 | |
| Minimum weight | −wmax/2 = −25 | |
| Eligibility trace | c1 = 1/wmax = 0.02 | |
| constants | c2 = 1 | |
| Maximum output | ymax = 5 |