Online Prediction of Multiple Tasks and the Netflix Prize

Ofer Dekel

Hebrew University

The simplicity and elegance of online learning make it a practical tool with many useful applications. We consider the situation where we are faced with multiple online prediction tasks in parallel, and where these tasks all contribute to a common goal. Instead of learning each task independently, our approach benefits from learning the multiple tasks jointly. We present two new online prediction algorithms for this problem and discuss an application of our technique to the "Netflix Prize" movie recommendation problem.