Intelligent Vision System-based Malware Classification using Deep Learning

  • Category: Computer Vision
  • Mentor: Prof. Priyanka Bagade
  • Designation: Dept. of CSE, IIT Kanpur
  • Period: Jan'22-Apr'22
  • Project URL: Will Update Soon

About

  • Designed an enhanced malware classification framework employing deep transfer learning to train directly on malware images. It provides a unique perspective for malware classification without dissembling the code, executing it, or extracting the features.
  • Addressed 3 problems -
    • Limited Data - There is minimal data for most notorious malware families such as Ransomware.
    • Imbalanced Dataset - Imbalance data may lead to overfitting the model or results as a biased model.
    • Lack of model generalizability - Several variants of different malware families introduced each day.
  • Few shots learning solved the first problem and second problem was solved by modifying log loss function which incorporates the different weights distributions for each class. Finally, third problem was solved using Meta-Learning that works best for out of distributions data that helped in domain generalization.