A meta-survey of adversarial attacks against artificial intelligence algorithms, including diffusion models
Marek Pawlicki, Aleksandra Pawlicka, Rafał Kozik, Michał Choraś
Abstract
Deep neural networks have revolutionized artificial intelligence, solving complex issues in areas like healthcare or law enforcement and security. However, they are susceptible to adversarial attacks where small data manipulations can compromise system reliability and security. This paper conducts an umbrella review of the literature on these attacks, synthesizing results from various systematic reviews to assess attack strategies, defense effectiveness, and research gaps. Guided by the PICO framework, this review categorizes and examines adversarial attacks, identifying key challenges in the field. The review finds that even though adversarial vulnerabilities were first explored in computer vision, analogous threats have expanded to domains like graph neural networks, natural language processing, federated learning, and text-to-image models. Despite varied attack surfaces, commonalities can be found. • First umbrella review synthesising systematic reviews and meta-analyses of adversarial attacks on deep neural networks, including the emerging threat to diffusion-based generative models. • PICO-driven framework addressing three research questions: (1) mapping survey themes and methods, (2) comparing domain-specific attack strategies, (3) identifying universal adversarial characteristics. • Comprehensive taxonomy covering gradient-based, transfer-based, score-based, decision-based, black-box, poisoning, privacy, and universal adversarial attacks. • Domain-specific analysis across computer vision, natural language processing, graph neural networks, intrusion detection systems, federated learning, GANs/VAEs, and text-to-image models like Stable Diffusion.