Machine & Deep Learning

AlexNet in 2012 started a new era in computing where deep learning started to find applications in every conceivable domain in science and engineering. This transformation was ignited by decades-long advances in high-performance computing and with the emergence of GPU's which made training deep neural networks practically feasible.

However, since 2012, both HPC and DL have been driving one another. On the one hand, HPC is pushing the limits of DL model complexity and depth of the neural networks for increased model accuracy while simultaneously reducing training time. On the other hand, DL is transforming the way we do science. The quest for knowledge used to begin with grand theories and physics-based models form the core of our understanding of science. However, data-driven models are beginning to outperform physics-based models in specific tasks although several challenges such as model interpretability and generalization remain to be addressed before such models become commonplace.

This page summarizes our research in accelerating machine and deep learning algorithms using parallel computing as well as applying DL techniques for accelerating scientific computing applications.

This project is partially supported by an NSF XPS award and NSF EAGER award.