Active Learning with Statistical Models-英文文献.pdf

MASSACHUSETTS INSTITUTE OF TECHNOLOGY ARTIFICIAL INTELLIGENCE LABORATORY and CENTER FOR BIOLOGICAL AND COMPUTATIONAL LEARNING DEPARTMENT OF BRAIN AND COGNITIVE SCIENCES A.I. Memo No. 1522 January 9, 1995 C.B.C.L. Pap er No. 110 Active Learning with Statistical Mo dels David A. Cohn, Zoubin Ghahramani, and Michael I. Jordan cohn@, zoubin@, jordan@ This publication can b e retrieved by anonymous ftp to . Abstract For many typ es of learners one can compute the statistically \optimal way to select data. We review how these techniques have b een used with feedforward neural networks [MacKay, 1992; Cohn, 1994]. We then show how the same principles may b e used to select data for two alternative, statistically-based learning architectures: mixtures of Gaussians and lo cally weighted regression. While the techniques for neural networks are exp ensive and approximate, the techniques for mixtures of Gaussians and lo cally weighted regression are b oth ecient and accurate. c Copyright Massachusetts Institute of Technology, 1995 This rep ort describ es research done at the Center for Biological and Computational Learning and the Articial Intelligence Lab oratory of the Massachusetts Institute of Technology. Supp ort for the Center is provided in part by a grant from the National Science Foundation under contr