README

Release Date

2018/07/23 It is uploaded after the deadline.

before

before

after adding adversal input

after

How to run source files

Problem1

Please Check Array.py

Problem2

run python problem2.py
it will print the Accuracy Percentage on your terminal

Problem3

run python problem3.py for eps0=0.1 AND baseline perturbation and save image files

run python problem3_report.py for eps0 = [0.001, 0.005, 0.01, 0.05, 0.1, 0.2, 0.5]

Problem4

run python problem4_1.py for FGSM Loop (Problem4.1)
I set 10 loops for xhat = x + eps
It will print out Accuracy for each Loop

run python prolbem4_3.py for DEFENDING from FGSM
run python prolbem4_3_attack.py for ATTACKING Modified Predictive Model

Source File Explanation

Array.py: NumPy Like Vector and Matrix(Limited) Calculation
layers.py: Relu, Affine, Softmax Layers are defined
functions.py: vector calculations such as softmax() are defined
fileloader.py: load and save pgm & Parameters
PretrainedThreeLayerNet.py: Three Layer Net with Relu, Affine, Softmax

REPORT

Please Check report.md in the same directory