Pattern Recognition:

Local Polynomial Models:

Local Polynomial Models have recently become a very popular method for regression and density estimation in the statistics literature. The following paper discusses how they may be used for pattern classification:

k-Nearest-Neighbor Algorithm:

Much of my work centers around the k-nearest-neighbor classifier. The fastest implementation of this algorithm that I know of, is described in

The finite sample risk of the k-nearest-neighbor classifier with different metrics is described in

Related work is contained in:

Empirical Estimation of the Bayes Risk:

The result of the finite sample analysis can be used to construct a statistical estimator of the Bayes risk. These studies are described in


Neural Networks:

An early study on pruning neural networks to obtain more accurate generalation for approximating continuous functions is described in


Physics:



Robert Snapp's Home Page
Updated last on March 10, 1998.