On-line learning with linear loss constraints

Abstract

We consider a generalization of the mistake-bound model (for learning {0, 1}-valued functions) in which the learner must satisfy a general constraint on the number M+ of incorrect 1 predictions and the number M- of incorrect 0 predictions. We describe a general-purpose optimal algorithm for our formulation of this problem. We describe several applications of our general results, involving situations in which the learner wishes to satisfy linear inequalities in M+ and M-. © 2000 Academic Press.