Estimating Classifier Generalization and Action's Effect: A Minimalist Approach

摘要

We present an online technique to estimate the generality of classifiers conditions. We show that this technique can be extended to gather some basic information about the effect of classifier actions in the environment. The approach we present is minimalist in that it is aimed at obtaining as much information as possible from online experience, with as few modifications as possible to the classifier structure. Because of its plainness, the method we propose can be applied virtually to any classifier system model.