Skip to content

arjn2/megan-aics

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

59 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

megan-aics

Enhanced Style-GAN based training model which extends dataset by generating new variants without compromising quality of generation.

Combined encoders with styleGAN for quality generation for new images.

Dataset used

Malevis balanced dataset from kaggle is modified with new class of additionally added benign pe images.

This custom benign class will reduce false positives.

dataset/pe2img.py is the python script used to convert pe files to gray images.

Optimizations

Well optimized for both gpu and cpu ,can run easily on google colab and local system also.

Strict paramters for memory efficiency, Avoid overfitting using early stopping patience, batch file optimizations ,etc.

improved training speed on GPU.

run well with 7-8 gb vram lock.

Improvement using CONFUSION MATRIX

The below show comparison of normal real dataset vs mixed dataset (real+generated) both are trained compared using cnn.

Mixed (Real+generated data) MODEL GAN-model

MODEL trained only in Real dataset NON-GAN

References

https://github.com/arjn2/MalDICT-reference
Dikedataset benign github

About

Improving malware detection using Style GAN

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published