Table 1 Comparison of the recognition accuracy on the MNIST dataset as a function of the number of columns, standard deviation of log (HRS) and log (LRS), and stuck-at-fault (%).
# of columns | Standard deviation (σ) | Probability of stuck-at-short (%) | Recognition accuracy (%) | |
|---|---|---|---|---|
w/o boost factor adjustment | w/ boost factor adjustment | |||
256 | 1 | 0 | 77.9 | 77.9 |
0 | 10 | 40.6 | 76.57 | |
1 | 10 | 37.4 | 76.57 | |
1,024 | 1 | 0 | 92.6 | 92.6 |
0 | 10 | 55.3 | 91.3 | |
1 | 10 | 51.2 | 91.1 | |
4,096 | 1 | 0 | 95.4 | 95.4 |
0 | 10 | 64.3 | 95.2 | |
1 | 10 | 61.2 | 94.5 | |
