Systems for Machine Learning

3 credits · Graduate

Principles of designing and implementing systems that enable running machine learning workloads efficiently on modern accelerators (e.g., GPUs). Topics include automated differentiation and backpropagation, machine learning compilers, GPU kernel design and optimization, distributed training, efficient inference techniques, emerging model architectures, parameter-efficient fine-tuning, and reinforcement learning. Prerequisites: Computer Science 310 or 510 or 512; and Computer Science 372 or 527 or 571 or 572 or 574 or 676.