Deep Learning @ UGA

From A/B testing to Markov and Bayes

In this talk we will review the continuum from A/B testing to contextual multi-armed bandits, to markov decision processes, and ultimately deep reinforcement learning. One of the challenges faced by reinforcement learning is the need for large amount of data for proper training. We will explore how probabilistic programming (Bayesian inference) can help with generating powerful environment models that can in turn enable training when only little, eventually non-stationary, and noisy data are available.

Dr. Stephane Pinel is a Sr. Manager of Data Science and Solution Architect. He has worked on the development of large scale recommendation engines, search engines and delivery via API at Cox Automotive, Merchandise and Financial Planning (MFP) solutions, supply chain management management solutions for large retailer (e.g. ToysRus, Walgreens, Best Buy...). He is currently managing 2 Data Science teams at MailChimp in the areas of look-alike-audience expansion and customer engagement metrics such as customer life value. His current research includes Artificial Intelligence and Automated decision making via a combination of Deep Reinforcement Learning and Probabilistic Programming (deep bayesian modeling, variational inference) as well as serverless microservice architecture for Data Science, “Big Data” data flow cloud architecture (AWS EMR/Spark, AWS lambda, AWS Kinesis, AWS API gateway, Data Pipeline…)). From 2006 to 2010, he was the CTO and Co-founder of Sayana Wireless, a startup company developing 60GHz multi-gigabit low power low cost radio for Consumer Electronics WPAN applications such as Wireless HDMI. In 2000, he received a Ph.D. in microelectronics and microsystems. He has published over 170 journals/proceeding papers,8 patents, 29 invention disclosure, over 3200 Google Scholar citations, and contributed to 2 wireless standards (and was vice-chair of ECMA committee) and book chapters. He gave numerous invited talks and organized numerous workshops at international conferences, managed over 30 PhD student, postdoc and research engineers and numerous industrial collaboration projects. His research interests included Advanced semiconductor devices modelling, millimeter-waves radio circuits and electromagnetic designs, Neural Network and Genetic Algorithms based designs, advanced 3D RF and millimeter-waves integration and antenna packaging technologies, RF-MEMS and micro-machining techniques.