Taking advantage of Goal Arity in Learning by Interaction

There exist many problems in the Reinforcement Learning domain which can be analyzed in terms of the so called goal arity. Once arity is determined, a general-purpose tabular learning approach can be applied with minimal memory requirements and attractive convergence properties. I will give examples of how this approach can be used.


Digital Media Initiatives and Research Opportunities at Amazon.com

In the course of running its many websites Amazon.com collects enormous quantities of information, such as session click-streams, purchase histories, product reviews, wish lists, etc. Additionally, the Digital Media Technology group ingests all sorts of digital media offered for sale, e.g., scanned books or movie/TV shows. This data presents great challenges but also great research opportunities. I will review the data available, show what we currently do with it, and discuss future possibilities.