Our research spans several cutting-edge domains, focusing primarily on the development and integration of intelligent systems for advanced AI applications. These systems include AI accelerators tailored for convolutional neural network (CNN), spiking neural network (SNN), Transformer, and AI-generated content (AIGC), aiming to enhance computational and energy efficiency as well as processing capabilities in real-world applications.
Furthermore, given the opaque nature of modern AI technologies, which lack transparency and unavoidably would raise concerns about trustworthiness and security issues, our work also seeks to address these issues with a primary focus on enhancing security during the inference phase, exploring the root causes of unintended behaviors, and developing holistic security solutions through software and hardware codesigns. It is expected to obtain a deeper understanding of AI/ML system vulnerabilities and to create novel holistic security solutions against various vulnerabilities such as adversarial attacks and data manipulation.
AI Accelerator Design
Hardware Security and Trustworthy Computing
