Session

DescriptionThis session charts a multifaceted exploration of how to optimize AI system efficiency. Presentations will explore dynamic voltage and frequency scaling for deep neural networks, dissect server overheads during DNN inference, and describe the use of active learning for Design-Technology Co-optimization (DTCO). The session will additionally address optimization of tensor programs, analytical cost-models for cross-platform performance prediction, advanced graph representation learning for performance modeling, and accelerating training of physics-informed neural networks.