Implementation of Guilding Deep Learning System Testing Using Suprise Adequacy https://arxiv.org/abs/1808.08444

In folder TEST, LSA,DSA，and SC can be run in a very small dataset of MNIST (100 training data and one new input).

In folder SA, I will try to modulize the codes and expand one new input to multiple new input.

TO DO:

SA/LSA.py

Fix the glitch when compute KDE in TEST/MNIST-LSA-TEST.py