Asheber Abebe
Department of Mathematics and Statistics
Auburn University
We will discuss robust signed-rank estimation of the parameters of a general nonlinear regression model where robustness is achieved through down-weighting in the gradient space. We use classic Sobolev space theory to help establish the asymptotic properties of the proposed estimator. We will discuss extensions to multidimensionally indexed nonlinear models including harmonic type functions used in texture modeling and signal processing. These functions do not typically satisfy certain Lipschitz conditions assumed in classical nonlinear regression modeling. So, asymptotic results are established under weaker conditions.
All statistics students are expected to attend.