Table 3.
SQNR and QNN accuracy for MLP trained on MNIST dataset with the application of different two-bit UQ designs [34].
wmin = −7.063787 wmax = 4.8371024 xmax[H] = 1.9605 xmax[J] = 2.1748 |
Case 1 ℜg [−wmax, wmax] |
Case 2 ℜg [wmin, −wmin] |
Case 3 ℜg [−xmax[H], xmax[H]] |
Case 4 ℜg [−xmax[J], xmax[J]] |
---|---|---|---|---|
SQNRexUQ(dB) | 2.8821 | −1.2402 | 8.7676 | 8.7639 |
SQNRthUQ(dB) | 1.9360 | −2.0066 | 6.9787 | 7.0707 |
Accuracy (%) | 96.97 | 94.58 | 96.34 | 96.74 |
Within ℜg (%) | 99.988 | 100 | 94.787 | 96.691 |