In recent years, rapid advancements in artificial intelligence (AI) have led to transformative changes in various industries, offering opportunities to automate tasks, enhance decision-making, and foster innovation. One example is the ResNet [1] architecture, a Convolutional Neural Network model used widely in computer vision tasks such as image classification, object detection, and semantic segmentation. Similarly, the Transformer [2] model is widely used in natural language processing. However, with the increasing use of AI in collecting and processing vast amounts of personal data, such as biometric data, online activity, and location data, there is a growing concern for privacy and security vulnerabilities exploited by side-channel attacks.

This project aims to investigate and mitigate potential security vulnerabilities that target side-channel information in current AI technologies. The project consists of two parts: vulnerability investigation and mitigation solutions. In the first part, we plan to implement and deploy adversarial approaches that leverage various side-channel information, including timing, power consumption, and hardware performance events. In the second part, we will determine the root cause of these vulnerabilities and propose mitigation solutions to patch and prevent these vulnerabilities from being exploited by potential attackers. By addressing these privacy and security concerns, we can ensure that AI technologies continue to offer benefits to society while have been safeguarded potential risks.