On the Performance of Generative Adversarial Network by Limiting Mode Collapse for Malware Detection Systems.

Area of Science:

• Machine Learning
• Cybersecurity
• Artificial Intelligence

Background:

• Generative adversarial networks (GANs) are powerful machine learning models comprising a generator and discriminator.
• GANs face common failure modes, notably mode collapse, which limits their ability to generate diverse and novel data.
• Mode collapse in GANs leads to instabilities and reduced performance in generative tasks.

Purpose of the Study:

• To investigate and address mode collapse in the Intrusion Detection System (IDS) Control Flow GAN (ICF-GAN) model.
• To propose conditional limiter solutions to mitigate mode collapse instances within the ICF-GAN framework.
• To enhance the accuracy and stability of GAN-based Intrusion Detection Systems.

Main Methods:

• The study focuses on conditional limiter solutions for mode collapse.
• A specific mini-batch method is employed to constrain mode collapse in the ICF-GAN.
• Performance is evaluated through experimental numerical results.

Main Results:

• The implemented mini-batch method effectively limits mode collapse instances in the ICF-GAN.
• The proposed solution significantly improves the accuracy of the Intrusion Detection System model.
• Experimental evaluations confirm the efficacy of the conditional limiter approach.

Conclusions:

• Conditional limiter solutions, particularly the mini-batch method, are effective in addressing mode collapse in ICF-GANs.
• The improved ICF-GAN model demonstrates enhanced accuracy for Intrusion Detection Systems.
• This research contributes to more stable and reliable GAN applications in cybersecurity.