Swetha Mandava (NVIDIA)


Swetha Mandava is a Senior Deep Learning engineer at NVIDIA where she develops optimized deep learning algorithms for applications in NLP/CV. She received her M.S in Electrical and Computer Engineering focusing on Machine learning from Carnegie Mellon University.


Distributed Large Batch Training

[slides, video]

Abstract: With increasingly complex Deep Learning models and datasets, AI practitioners are faced with escalating training times, and hence lower productivity. In this workshop, we will scale a prototype to production quality in 90 minutes. Starting with a popular recommender system we will explore convergence, stability, and scaling techniques to drastically improve performance and reduce training time by about 35x.