Robustness_analysis
│   README.md
│   synaptic_filtering.ipynb
|   model_trainer.ipynb
|   plots
│
└───Robustness_analysis
│   └───Dataset
|   |   └───MNIST
|   |   |   └───processed
|   |   |   |   └───test.pt
|   |   |   |   └───training.pt
|   |   |   └───raw
|   |   |   |   └───t10k-images-idx3-ubyte
|   |   |   |   └───t10k-images-idx3-ubyte.gz
|   |   |   |   └───t10k-labels-idx3-ubyte
|   |   |   |   └───t10k-labels-idx3-ubyte.gz
|   |   |   |   └───train-images-idx3-ubyte
|   |   |   |   └───train-images-idx3-ubyte.gz
|   |   |   |   └───train-labels-idx3-ubyte
|   |   |   |   └───train-labels-idx3-ubyte.gz


|   |   └───CIFAR10
|   |   |   └───cifar-10-python.tar.gz
|   |   |   └───cifar-10-batches-py
|   |   |   |   └───batches.meta
|   |   |   |   └───data_batch_1
|   |   |   |   └───data_batch_2
|   |   |   |   └───data_batch_3
|   |   |   |   └───data_batch_4
|   |   |   |   └───data_batch_5
|   |   |   |   └───readme.html
|   |   |   |   └───test_batch

|   |   └───tiny-imagenet-200
|   |   |   └───complete_dataset
|   |   |   |   └───test_set.pkl
|   |   |   |   └───train_set.pkl
|   └───saved_models
|   |   └───ResNet18
|   |   |   └───MNIST
|   |   |   |   └───Init_1
|   |   |   |   └───...
|   |   |   |   └───Init_N
|   |   |   |   └───global_data
|   |   |   |   └───Step_minmax
|   |   |   |   └───Step_maxmin
|   |   |   |   └───Pulse_minmax

|   |   |   └───CIFAR10
|   |   |   └───ImageNet
|   |   |   |   └───



## Running code

To run the experiments, open the .ipynb from root directory and follow cell comments. 

## links to datasets (open source dataset)

MNIST - http://yann.lecun.com/exdb/mnist/
CIFAR10 - https://www.cs.toronto.edu/~kriz/cifar.html
ImageNet Tiny - https://www.kaggle.com/c/tiny-imagenet


## What are the files and what are their purpose

1) The networks first need to be trained and saved into different Init folders within Saved_model->Network_name->dataset_name->Init_i. the networks can eb trained using the model_trainer.ipynb notebook, cell instructions are provideed within the notebook.

2) After the networks have been trained and svaed within the correct directories, it ispossible to run the synaptic filtering expeirments from the Synaptic_filtering.ipynb notebook. Here there are options for global and local anlaysis.

## Dependencies

To train and run the synaptic filtering experimetns on all networks, the following dependencies are required:

import matplotlib.pyplot as plt
from matplotlib.ticker import MaxNLocator, FormatStrFormatter, AutoMinorLocator
from tqdm.notebook import tqdm
import copy
from mpl_toolkits.mplot3d import Axes3D
import numpy as np

import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import DataLoader
from torch.autograd import Variable

import torch_xla
import torch_xla.core.xla_model as xm
import torch_xla.distributed.xla_multiprocessing as xmp
import torch_xla.distributed.parallel_loader as pll
import torch_xla.utils.serialization as xser

import torchvision.transforms as transforms
import torchvision.datasets

import random
import os
import pickle