Clustering Multiple Curves

My primary focus in the analysis of multiple curves is to cluster or classify curves based on similar behavior over time or cluster by shape. In my research related to this methodological area, I investigated methods under different assumptions starting with independence between constant-variance curves and extending this common assumption to error heteroskedasticity and spatial-dependence between curves. The cluster techniques that I have published allow for a large number of curves observed at a small number of time points and with low contrast-to-noise ratio, a difficult statistical setting.

Papers:
1. Serban, N., Wasserman, L. (2005) "CATS: Cluster Analysis by Transformation and Smoothing", , Journal of the American Statistical Association, 100. [.pdf ]
2. Serban, N. (2008), "Clustering Curves in the Presence of Heteroscedastic Errors", Journal of Nonparametric Statistics. [.pdf]
3. Serban, N. (2008), "Clustering Confidence Sets", Journal of Statistical Planning and Inference. [.pdf]
4. Jiang, H., Serban, N. (submitted 2008), "Large Scale Clustering of Dependent Curves".[.pdf]