Improved Risk Tail Bounds for On-Line Algorithms

We prove the strongest known bound for the risk of hypotheses selected from the ensemble generated by running a learning algorithm incrementally on the training data. Our result is based on proof techniques that are remarkably different from the standard risk analysis based on uniform convergence arguments.

• Type: Journal
• Year: 2008
• Where: TIT
• Authors: Nicolò Cesa-Bianchi, Claudio Gentile
• Keywords: Proof Techniques, Standard Risk Analysis, TIT 2008, Uniform Convergence Arguments