DEEP LEARNING SYSTEMS DESIGN AND OPTIMIZATION

Credit Hours: 3 · Undergraduate Upper-Level

Deep learning has transformed scientific discovery, industry, and everyday applications. However, the success of deep learning relies critically on the systems infrastructure—GPUs, accelerators, distributed frameworks, compilers, and runtime systems—that enable scalable training and efficient inference. This course provides a deep dive into the principles, design, and optimizations of deep learning systems. Topics include GPU architecture and programming for deep learning, parallel and distributed training strategies, memory management and scheduling, compiler techniques for deep learning frameworks, systems support for large models (e.g., LLMs and diffusion models), and co-design across algorithms, systems, and hardware. Students will gain both theoretical foundations and hands-on experience through assignments and a semester-long project, where they design and evaluate systems optimizations for real-world deep learning workloads. Graduate/Undergraduate Equivalency: COMP 568 . Recommended Prerequisite(s): COMP 330 or COMP 341 Mutually Exclusive: Cannot register for COMP 468 if student has credit for COMP 568 .