Computer Science Seminar by Yue Dai: Towards Efficient and Robust Graph Deep Learning聽
Speaker: , Ph.D. student, University of Pittsburgh
Title: Towards Efficient and Robust Graph Deep Learning
Abstract:
Inspired by the success of Graph Neural Networks (GNNs), recent graph deep learning studies have introduced GNN-based models like Graph Matching Networks (GMNs) and Temporal Graph Neural Networks (TGNNs) for diverse tasks in various domains such as social media, chemistry, and cybersecurity. Despite these advances, deploying such models efficiently and robustly in real-world settings remains challenging. Three core issues impede their broader adoption: (1) suboptimal inference latencies, which fail to meet real-world responsiveness needs; (2) limited training efficiency and scalability, hindering rapid model development for targeted applications; and (3) fragile robustness against adversarial attacks, posing serious security and privacy concerns.
This talk will present my research on full-stack optimizations for GNN-based models. First, I will introduce Cascade, a dependency-aware TGNN training framework that boosts training parallelism without compromising vital graph dependencies, resulting in faster training while preserving model accuracy. Next, I will detail CEGMA, a software-hardware co-design accelerator that eliminates redundant computations and memory accesses in GMNs, which leads to faster inference. Finally, I will outline future directions that push the frontier of deep graph learning: optimizing efficiency in emerging GNN鈥揕LM hybrid models, investigating the adversarial vulnerabilities of TGNNs alongside robust defenses, and applying graph-based reinforcement learning to tackle system design challenges. Through this holistic approach, I aim to enable efficient, scalable, and secure GNN-based solutions across a wide range of real-world applications.
Bio:
Yue Dai is a final-year Ph.D. student in the Department of Computer Science at the University of Pittsburgh, advised by Dr. Youtao Zhang and Dr. Xulong Tang. His research focuses on machine learning systems and accelerators, with an emphasis on efficient systems and robust algorithms for deep graph learning. Specifically, he explores techniques to accelerate Graph Neural Network-based models across diverse platforms, including GPUs and ASICs, and to enhance their robustness against adversarial attacks. His work has been published in top-tier venues across computer architecture and systems (e.g., ASPLOS, HPCA) as well as machine learning (e.g., ICLR).