Learning switching concepts

We consider learning in situations where the function used to classify examples may switch back and forth between a small number of different concepts during the course of learning. We examine several models for such situations: oblivious models in which switches are made independent of the selection of examples, and more adversarial models in which a single adversary controls both the concept switches and example selections. We show relationships between the more benign models and the p-concepts of Kearns and Schapire, and present polynomial-time algorithms for learning switches between two k-DNF formulas. For the most adversarial model, we present a model of success patterned after the popular competitive analysis used in studying on-line algorithms. We describe a randomized query algorithm for such adversarial switches between two monotone disjunctions that is “1-competitive” in that the total number of mistakes plus queries is with high probability bounded by the number of switches plus some fixed polynomial in n (the number of variables). We also use notions described here to provide sufficient conditions under which learning a p-concept class “with a decision rule” implies being able to learn the class “with a model of probability.”.

[1]  Ming Li,et al.  Learning in the presence of malicious errors , 1993, STOC '88.

[2]  Ronitt Rubinfeld,et al.  Reconstructing algebraic functions from mixed data , 1992, Proceedings., 33rd Annual Symposium on Foundations of Computer Science.

[3]  Lyle A. McGeoch,et al.  Competitive Algorithms for Server Problems , 1990, J. Algorithms.

[4]  Ronald L. Rivest,et al.  Learning decision lists , 2004, Machine Learning.

[5]  Philip M. Long,et al.  Tracking drifting concepts using random examples , 1991, Annual Conference Computational Learning Theory.

[6]  Linda Sellie,et al.  Toward efficient agnostic learning , 1992, COLT '92.

[7]  Robert H. Sloan,et al.  Corrigendum to types of noise in data for concept learning , 1988, COLT '92.

[8]  R. Schapire Toward Eecient Agnostic Learning , 1992 .

[9]  Allan Borodin,et al.  An optimal online algorithm for metrical task systems , 1987, STOC.

[10]  Robert E. Schapire,et al.  Efficient distribution-free learning of probabilistic concepts , 1990, Proceedings [1990] 31st Annual Symposium on Foundations of Computer Science.

[11]  Lyle A. McGeoch,et al.  Competitive algorithms for on-line problems , 1988, STOC '88.

[12]  Allan Borodin,et al.  On the power of randomization in online algorithms , 1990, STOC '90.

[13]  Esther Levin Modeling Time Varying Systems Using Hidden Control Neural Architecture , 1990, NIPS.

[14]  Robert E. Schapire,et al.  Efficient distribution-free learning of probabilistic concepts , 1990, Proceedings [1990] 31st Annual Symposium on Foundations of Computer Science.