Step into the world of resilient AI models! Our thesis tackles the growing threat of cyber-attacks on Artificial Intelligence systems, particularly in the realm of time-series AI. As AI becomes integral to critical processes, we explore how cyber threats can target prediction and decision-making, risking privacy, accessibility, and integrity. This project conducts a detailed analysis of vulnerabilities and threats, studies the impact on specific applications, and proposes practical defenses. Join us in enhancing AI resilience, employing techniques like adversarial training to fortify against evolving cyber challenges. It's a journey into securing the future of smart technology!
The goal of this work is to apply adversarial training methods on an industrial control system dataset, in order to increase resilience of time series AI. First experiments should be done using publically available datasets and state-of-the-art time series methods, while later on more challenging data readings will be provided in the project.