Skip to content

Using Julia to generate adversarial examples to machine learning models

Notifications You must be signed in to change notification settings

Pgibby8/Adversarial-Julia

Repository files navigation

Adversarial Julia

Inventory

Adversarial.jl: the file containing the adversarial module
Demo.ipynb: a notebook to demonstrate how the module works
Midterm Progress.ipynb: A notebook that includes legacy results. The first few cells deal with generating adversarial examples against a malware classifier
Project Presentation.pptx: The presentation I gave in class

Required Packages

Distances, Knet, StatsBase, Optim, AutoGrad, ProgressMeter, Statistics, ForwardDiff, IterTools

Running the Code

See the "Demo" notebook for examples. In general, first load the data, initialize a classifier (this can be done with the built in adv.MLP struct), and train it (with adv.trainresults). Once trained, you can run an attack on the model using one of the attack functions defined in Adversarial.jl. Use adv.targetted_attack and adv_untargetted_attack to perform an attack over all input

Malware Classification Attack

This code can be found in the first several cells of Midterm Progress.ipynb

About

Using Julia to generate adversarial examples to machine learning models

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published