Clustering with Outliers Samir Khuller University of Maryland Abstract: Standard clustering measures such as K-center, K-medians and facility location (where there is a cost for creating a cluster) require a clustering in which every single input point has to be clustered. In some cases (for example with noisy data) we may obtain better clusterings if we are allowed to remove a small number of outliers. In this survey talk, we focus on the K-center problem and discuss several basic results in the context of clustering with outliers. We develop approximation algorithms for both the standard computation model when the entire input data is available in advance, and also algorithms for maintaining a clustering in the streaming model. We briefly discuss methods for facility location as well. (The first paper is joint work with M. Charikar, D. Mount and G. Narasimhan and the second is joint work with R. McCutchen.)