Sequential Prediction, Martingale Tail Bounds and Automatic Machine Learning

• Activity Number: 602 - Theory at the Intersection of Machine Learning and Statistics
• Type: Invited
• Date/Time: Thursday, August 2, 2018 : 8:30 AM to 10:20 AM
• Sponsor: IMS
• Author(s): Karthik Sridharan*
• Companies: Cornell University
• Keywords: Sequential prediction; tail bounds; empirical process theory

Abstract

At the get go, sequential prediction, uniform martingale tail bounds and convex/concave analysis don't seem obviously related. But in this presentation we will see how the concepts are inherently interlinked. Specifically we will see how the online prediction framework for sequential predictions is inherently tied to uniform martingale tail bounds. In fact using this connection we will see how one can obtain martingale concentrations in Banach spaces. We will also see that this connection will yield a simple and elegant way for designing adaptive machine learning algorithms. Specifically, we will use a classic result of Donald Burkholder one the so called zig-zag concave functions to design efficient and elegant estimators for prediction problems that harness the connection between sequential prediction and probabilistic inequalities. We will use this insight to help us get a step closer to what I shall term Plug-&-Play ML. That is, help us move a step towards building machine learning systems automatically.