Probabilistic Evaluation of Bias for Learning Systems

Marie desJardins
Machine Learning Workshop, Evanston, Illinois. Morgan Kaufmann, 1991.

A method for using probabilistic background knowledge to select features for representing learned concepts is described. The method uses a model of the tradeoff between accuracy (predictiveness) and simplicity (size) of hypothesis spaces. Preliminary results of using various biases for a learning task in a randomized test domain are discussed.

Click to get a postscript version of this paper.

Marie desJardins, mariedj@cs.umbc.edu