First Data Science SIG of the year! We have two Data Scientists Mark Hoffmann and Zax Rosenberg talking about their projects.
Talk 1: Neural Networks for Multi Level Forecasting
Forecasting is a hard problem to do well. What happens when you are not only forecasting a single series, but you have to forecast tens of thousands of series simultaneously? This talk will cover some non parametric methodologies for working with this idea of many time series. We will go over basics of neural networks, entity embeddings for categorical features, competing methodologies, real world implementations that have attained state of the art results, and some common applications of these models such as anomaly detection.
About the Speaker:
Mark Hoffmann has been working on business and research endeavors related to technology for nearly a decade. These pursuits began at Augustana College where he worked in a research lab that focused on high energy nuclear physics as well as one that was focused on quantum optics. While in undergrad, he co-founded a software company called 38th Street Studios. Mark then ventured to Raleigh, North Carolina where he worked towards a masters in analytics from NC State's Institute for Advanced Analytics while working with a government agency on data science efforts related to network optimization and event sequencing.
Following development efforts of 38th's first major software platform, Mark moved back to Chicago to work full time on the startup, which later went on to act as a major transportation logistics system that has served executive travel for major events such as the 2018 Superbowl. Continuing to work towards a goal of democratizing technology, data, and automation in every day lives, he has helped lead efforts and develop software for clients of 38th Street Studios that continue to flourish with great people in domains that span logistics, construction, and real estate.
Mark has recently spent time in health insurance, where he has worked on teams to optimize provider networks as well as leading an initiative for a time series based anomaly detection framework.
