Mal_InEx_GAN

A Research on Malware (Malhub) classification with ExceptionNet and InceptionNet - creation of a GAN Model

Image Type (GRAY scale)

1. Model-1 (Inception Net) -> Use GAN -> Samples (Adv. example) from each test set -> drop the performance
2. Model-2 (Xception Net) -> Use GAN -> Samples (Adv. example) from each test set -> drop the performance

Note: Take same test examples (eg: M1--> Model1 & Model2)

3. CAM -> Training set for Model 1 and Model 2
   
4. CAM -> Test samples before using GAN and after using GAN

Show

• CAM of the test of each family is identical to training set, but the modified instance of each family have distinct pattern

5. Use packer -> observe the drop in detection -> see the CAM
6. If we augment the dataset using GAN generated examples, performance improves
   

• some % of samples (traning samples) -> GAN -> new instance -> extend the dataset -> model1 & model2
  
• test set (old) -> present to model1 & model2 -> we observe that test set samples which were previously misclassified are also detected

B E F O R E   A T T A C K

The Bellow image is the depiction of classification report before the adversarial attack.

Classification Report



Performance Metrics

GE N E R A T I N G   S A M P L E S