Abstract: While deep learning models may outperform traditional counterparts in computer vision and language tasks, deploying these algorithms at scale is challenging. Online resources focus too heavily on cool new architectures, model training, and ousting performance benchmarks, whilst lacking more practical tips and tricks for implementing deep learning solutions within your organization.
In this talk, we’ll walk through the details and lessons learned from developing and deploying a visual search service at ShopRunner. You’ll learn to leverage Docker to develop code that is agnostic to CPU and GPU compute environments, evaluate your models (yes! even unsupervised ones), design a scalable service architecture with low latency and high availability, utilize your org’s continuous integration and continuous deployment (CI/CD) pipelines, and communicate what you’ve done with the rest of your org.
You’ll get the most out of this session if you have some familiarity with deep learning, and have deployed code within a production stack. Though we’ll discuss code snippets in Python, familiarity with any programming language will suffice.
Bio: Scott Cronin is a Sr. Data Scientist at ShopRunner, where he builds and deploys predictive models at scale for over 140 online retailers and 6 million members. Prior to ShopRunner, he was a Sr. Data Scientist at Trunk Club, a technology-driven fashion company acquired by Nordstrom, and a Sr. Engineer at Intel, where he improved manufacturing technology for computer chips with test-driven statistical modeling and experimentation techniques. Scott earned a Ph.D. in Materials Engineering from Northwestern University where he developed and automated computer vision approaches to correlate fuel cell performance with microscopy images of a device's internal structure. He is an avid hiker, CrossFitter, and snowboarder who loves the outdoors. He currently resides in Salt Lake City, Utah