Tyler Whitehouse (hosted by CIS) – “Portable, reproducible but user friendly deep learning with Docker Machine, PyTorch and Gigantum”
“Portable, reproducible but user friendly deep learning with Docker Machine, PyTorch and Gigantum”
Abstract: This talk will demonstrate how to use Docker Machine, PyTorch and Gigantum to create and work with Jupyter notebooks to do deep learning. In particular, we will demonstrate how to use these three tools to create notebooks that can be easily moved between different resources, from laptop to remote, in order to work on both CPUs and GPUs. We will use a simple transfer learning tutorial to demonstrate how easy it is for anybody, even a novice, to get up and running in a short amount of time. Time permitting, we will also show how to do similar things using TensorFlow in the context of training an LSTM speech generator.
Bio: Tyler Whitehouse graduated from the University of Minnesota with a doctorate in mathematics in 2009, and continued on to a postdoc with the approximation theory group at Vanderbilt University until 2012. He then worked as a consultant in the DC area, assisting with developing and managing machine learning programs at agencies such as DARPA, IARPA and ARPA-E. Currently, he is the CEO of Gigantum, an early stage startup that is developing a data science platform.