The Need for Efficiency in Modern Deep Learning
As deep learning models become exponentially larger and more complex from millions to trillions of parameters—a significant challenge has arisen: conventional neural architecture design frequently overlooks the essential limitations of the hardware on which these models operate. This gap has caused models that are theoretically strong but not very useful in practice, using too much energy, memory, and computing power.

Hardware-Aware Neural Architecture Design and Optimization is a new way of doing AI research that combines algorithms and hardware to get the best performan