Nvidia Deep Learning Software and Optimizations
Brown Bag Lecture by Jesse Tetreault | 9/18/2018 11:00AM – 12PM | 7th Floor Conference Room, Bldg 38A
Deep learning has become the most prevalent form of Artificial Intelligence in recent years due to its success on a wide range of problems including Computer Vision, Natural Language Processing, Fraud Detection, Risk Analysis, and much more. This new wave of Deep Learning has been fueled by 3 things: more advanced neural network models, new datasets becoming available in many areas, and a significant advance in the hardware used to train these new models; namely, GPUs from Nvidia. In this talk we will discuss popular Deep Learning software frameworks such as Tensorflow and Pytorch and best practices around fully utilizing Nvidia hardware when deploying them. This will include discussion of mixed-precision (FP16) training on new Volta GPUs, tips and guidance around training multi-GPU deep learning models, and a survey of popular tools that Nvidia supports in many phases of a Deep Learning workflow. An additional key component of this talk will be discussion of how to best use containerized workflows such as Docker and Singularity in conjunction with Nvidia GPU Cloud (NGC) to take advantage of all the latest and greatest framework and hardware optimizations.
Jesse Tetreault has been a Solutions Architect at Nvidia since September 2017 specializing in Deep Learning and High-Performance Computing. Prior to joining Nvidia, Jesse received a M.S. degree in Computer Engineering from Clemson University in South Carolina. While in school his research focus was on applying Deep Learning to Computer Vision tasks, particularly combining multiple image modalities such as color and stereo images to improve semantic segmentation tasks. Since joining Nvidia, Jesse has worked with customers in many different verticals including Healthcare, Financial Services, and Security. Jesse became the main Nvidia technical contact for NIH in June 2018 and continues to act as a liaison for the NIH back to Nvidia.