Learning classifiers that perform probabilistic classification under computational resource constraints

Existing online machine learning techniques are often tuned to minimized computational resources, and are not able to utilize further resources when they are available. This is likely to result in sub-optimal learning performance in many online applications where the computational resources available will vary from time to time. This paper analyzes categories of computational resources and defines the problem of classification learning using varying and uncertain computational resources. An anytime classification algorithm AAODE is proposed. AAODE begins with the classification time and accuracy of naive Bayes and can increase classification accuracy when more time resource is given.