Table 1 The distribution of non-zero entries in the MNIST dataset
From: The optimally designed autoencoder network for compressed sensing
K | 0–50 | 51–100 | 101–150 | 151–200 | 201–250 | 251–784 | Total | Average |
|---|---|---|---|---|---|---|---|---|
Training set | 75 | 6661 | 21,204 | 20,821 | 5717 | 522 | 55,000 | 149.93 |
Validation set | 5 | 607 | 1918 | 1905 | 533 | 32 | 5000 | 149.63 |
Testing set | 13 | 1163 | 3770 | 3942 | 996 | 116 | 10,000 | 151.12 |