Dr. Fangtian Zhong (School of Computing, MSU)

03/13/2025 3:10pm

Abstract:

Machine learning enables computers to learn from experience and interpret the world through a hierarchical structure of concepts, where each concept is defined by its relationship to simpler ones. This has driven advancements in various domains, including language translation, image recognition, social media analytics, speech recognition, and malware detection. However, the growing reliance on machine learning has also introduced new vulnerabilities, allowing sophisticated adversaries to exploit these systems. Despite this, real-world adversarial malware examples capable of successfully bypassing modern anti-virus products remain scarce, as these products integrate machine learning with traditional detection techniques. Moreover, existing anti-virus solutions are not inherently designed to facilitate the automatic mitigation of malware-induced damage and handle packed samples. In this presentation, I will explore the application of machine learning to enhance the capabilities of malware in evading anti-virus detection while simultaneously developing more precise and storage friendly defense strategy to improve malware damage recovery.