Almost-everywhere algorithmic stability and generalization error

Samuel Kutin, Partha Niyogi

We explore in some detail the notion of algorithmic stability as a viable framework for analyzing the generalization error of learning algorithms. We introduce the new notion of tr…