Recently, Professor Zhu Heguai's research team from the Department of Mathematics, College of Sciences, NEU has made new progress in the field of adversarial attacks for AI security. The research paper, titled "Reinforcing Adversarial Transferability via Negative Class Guided Example Generation" has been accepted for publication in IEEE Transactions on Information Forensics and Security (IEEE TIFS), a top-tier journal in network and information security. Professor Zhu Heguai is the first author, and NEU is the primary affiliation.
This paper investigates adversarial sample generation from the perspective of deep neural network model decision-making, establishing a connection between targeted and non-targeted attacks. It proposes a decoupled non-target attack algorithm—Negative Class Guidance (NCG). NCG utilizes the outputs of proxy models to identify weak decision-making directions within deep neural network models. By introducing high-confidence negative samples, it provides superior update directions for adversarial examples, significantly enhancing their transferability across different model architectures. Additionally, NCG can be integrated with most existing adversarial attack methods targeting deep neural networks, demonstrating broad applicability. It reveals the vulnerabilities inherent in current deep neural network classification models, thereby providing essential support for the security protection and practical deployment of deep neural network models.
IEEE TIFS focuses on cutting-edge research in information forensics and security, AI security, cryptography, and privacy protection, encompassing theoretical research, algorithm design, and practical applications. It is one of the three Class A journals recommended by the China Computer Federation (CCF) in the field of network and information security (the other two being IEEE TDSC and Journal of Cryptology). It is also a Class A journal recommended by the Chinese Association for Cryptologic Research (CACR) and a top-tier journal in the Q1 category of the Chinese Academy of Sciences, enjoying high academic standing and significant industry influence within the field. This research has been supported by the National Natural Science Foundation of China.
