A project to inspect neuron features doing gradient descent w.r.t to the input

First it has been created a low level Neural Network class ('NN.py') implmented with pytorch, which gives the possiblity to inspect the activations of all the neurons.
With the possibility to get the activations of all the neurons it has been created a FeatureDetector class ('FeatureDetector.py') that perfrom Stochastic Gradient Descent w.r.t to the input in order to maximise the activation of a given neuron, by doing this we are able to retrieve which are the features which excite our neuron.
Note that with this tecniques it's easy to perform Adversiarial Attack on the Network and also to train it in a more robust way.

The code is written in python 3.6 by using the following packages:

• pandas
• numpy
• matplotlib
• cv2
• sklearn
• scipy
• pytorch
• time
• itertools