Open digantamisra98 opened 1 year ago
Hi @digantamisra98
Thanks for your question and pardon me for my delays with getting back.
For the top figures, Im just taking random 2D slices centered on the origin while changing the input space dimensionality. Its computed for an NN with a single hidden layer for which Im also varying the width.
For the bottom figures, I took 2D slices centered on 50 test samples from tiny-Imagenet, computed partition statistics for networks pre-trained on tiny-Imagenet and plotted the histogram. The networks were pre-trained both with and without data augmentation.
I'll make the codes available soon but am quite backlogged.
What would be the way to plot these two figures?