We prove that access to the prior distribution over target functions can improve the sample complexity of self-terminating active learning algorithms, so that it is always better than the known results for prior-dependent passive learning. In particular, this is in stark contrast to the analysis of prior-independent algorithms, where there are known simple learning problems for which no self-terminating algorithm can provide this guarantee for all priors.

14 pages

This page maintained by reports@cs.cmu.edu