Abstract: Deep learning is an area of machine learning that has become ubiquitous with artificial intelligence. PyTorch provides a comprehensive framework for the development of deep learning models. However, project requirements often extend beyond the model development process. SAS has a rich set of established and unique capabilities that support model development and deployment, including some new features that use the TorchScript language. In this workshop, we will demonstrate how to integrate PyTorch with SAS to leverage the benefits of both technologies. The workshop will focus on computer vision applications, but the framework can easily be extended to other deep learning tasks.
Participants will also learn how to improve model accuracy using combined global and local search strategies that are evaluated in parallel to ensure a quick and efficient exploration of the decision space. In the case of this workshop, a genetic algorithm will be used for the global search because the selection and crossover aspects of the genetic algorithm distinguish it from a purely random search. A generating set search will then be used to greedily search the local decision space.
Bio: Ari Zitin holds bachelor’s degrees in both physics and mathematics from UNC-Chapel Hill. His research focused on collecting and analyzing low energy physics data to better understand the neutrino. Ari taught introductory and advanced physics and scientific programming courses at UC-Berkeley while working on a master’s in physics with a focus on nonlinear dynamics. While at SAS, Ari has worked to develop courses that teach how to use Python code to control SAS analytical procedures.