Table 6:
Validation Errors at the End of the Training Period for a Fully Connected One-Hidden-Layer Network Trained on MNIST, with Different β Values.
| β value | Mean Validation Error (%) | Minimum Validation Error (%) | Maximum Validation Error (%) |
|---|---|---|---|
| 0.01 | 89.70 | 89.70 | 89.70 |
| 0.1 | 90.09 | 90.09 | 90.09 |
| 0.25 | 46.20 | 2.28 | 90.09 |
| 0.5 | 2.19 | 2.17 | 2.21 |
| 0.75 | 2.42 | 2.22 | 2.51 |
| 1.0 | 2.36 | 2.26 | 2.48 |
| 1.2 | 23.30 | 2.36 | 85.91 |
| 1.5 | 2.62 | 2.40 | 2.75 |
| 2.0 | 79.55 | 79.55 | 79.55 |
Notes: For values that lay within the parameter range that converged to low (< 3%) validation errors, four trials were run, and the mean, minimum, and maximum errors over the trials have been reported. In all runs, γ = 1.