Fast and Accurate Deep Neural Networks Training

Explore a colloquium presentation on fast and accurate deep neural networks training delivered by Yang You from UC Berkeley. Dive into the increasing popularity of supercomputers in leading AI companies and their potential for accelerating deep learning computations. Learn about the challenges of parallelizing deep learning and discover innovative optimizers like LARS and LAMB, which enable better scaling and higher accuracy in real-world applications. Gain insights into record-breaking Imagenet training speeds, reduced BERT training times, and state-of-the-art results on various benchmarks. Understand how these approaches are being utilized by major tech companies and their impact on distributed systems and supercomputers in the field of machine learning.