Fueled by machine learning (ML) model and hardware advancements, intelligence is transforming every walk of life. For critical applications like autonomous vehicles, ensuring inference dependability is essential. Unfortunately, current hardware cannot provide such a promise. This CAREER project aims to create a new paradigm of safeguarding ML execution against both passive hardware faults and active fault attacks. The novelties lie in the new capability development inside ML processing, and the cross-layer exploration of algorithm, architecture, and hardware security. The broader impacts include yielding practical solutions for ensuring the root of trust of accelerated intelligence services and abundant educational opportunities.

Connected and autonomous vehicles (CAVs) are set to revolutionize transportation systems, enhancing road safety and travel efficiency. Vehicle-to-Everything (V2X) network connectivity enables vehicles and infrastructure to exchange data, improving perception and decision-making. Cooperative perception introduces new CAV applications like cooperative driving, map updates, and platooning. Edge-assisted architectures, leveraging edge computing, offer solutions but face challenges like communication and computation latency, network dynamics, and heterogeneous capabilities. This project proposes a novel edge-assisted semantic communication framework, aiming for low-latency and reliable cooperative perception in safety-critical applications. It optimizes vehicle and edge data processing and V2X communication, advancing networking applications in line with the NextG vision.

Machine learning (ML) as a service on cloud is pervasive, but it poses real threats to personal or business providers' privacy. To guarantee privacy, cryptographic protocols, such as Homomorphic Encryption, are promising due to enabling ML analytics directly on encrypted data. However, there exists a big gap between the theory and practice, e.g., long latency due to the prohibitively expensive computation or communication overhead over ciphertext. This project aims to practically accelerate the private ML service by offering a full-fledged development of efficient, scalable and encryption-conscious computing paradigms. The broader impacts include advance trustworthy artificial intelligence and educational opportunities.

Future computer data centers face increasing demands for high-computation workloads driven by power-hungry deep neural network (DNN) models. DNN accelerators, employing processing in memory with new storage devices, promise energy efficiency and performance. However, these accelerators encounter stability challenges due to device limitations. This project aims to develop efficient neural network approaches to address this issue. The impact includes potent, scalable deep learning systems benefiting various applications and fields, such as business, science, and national security. It also elevates students' competence and confidence in the competitive job market, integrating research results into courses and outreach activities.