Skip to main content
. 2020 Jan 16;22(1):108. doi: 10.3390/e22010108

Figure 11.

Figure 11

Original MNIST labels. The top row shows lowest predictive label information among original labels. Notably, the two leading images appear to be genuinely mislabeled in the original dataset. Subsequent predictions are poor. The bottom row shows the highest information among original labels. Labels and predictions are consistent in these cases.