- Python 3.6+
- Pillow, matplotlib, numpy, pytorch==0.4.0, seaborn, torchvision
mnist_experiments.pyruns a full set of experiments on MNIST and save the results to the directory
saved. Note: the default run take a long time (43 hours) to finish, since we're running for all 10 seeds.
Currently, it executes the following experiments:
- Measure the difference between exact augmented objective and approximate objectives (on original images, 1st order approximation, 2nd order approximation).
- Measure the agreement and KL divergence between the predictions made by model trained on exact augmented objective and models trained on approximate objectives.
- Compute kernel target alignment for features from different transformations.
plot.pyplots all the figures in the paper using the saved results from
mnist_experiments.py. The figures are saved in the directory