In this	work, we show how to take advantage of parametric generative
models for learning local metrics to improve nearest neighbor
classification.  In particular, we show how finite sampling introduces
bias in the information-theoretic performance of a nearest neighbor
classifier, and how the optimal local metric can be analytically found
from a semi-definite programming problem.  We demonstrate how this
learned metric enhances the discriminative nearest neighbor
performance on various datasets using simple Gaussian class
conditional generative models.