Recent results on the analytic center approach for bounded error parameterestimation

In this paper, we present an overview of some recent work [5] on the so-cal led analytic center approach for bounded error parameter estimation. First, we discuss the optimality properties of well-known algorithms such as the Chebychev center, the projection and the min-max estimates. Subsequently, w e propose the analytic center as an alternative algorithm for recursive est imation. We show that the analytic center minimizes the output error and, o n the contrary of other estimates like Chebychev, allows for an easy-to-com pute sequential algorithm. We argue that the maximum number of Newton itera tions required to evaluate a sequence of analytic centers is linear in the number of observed data points and it is comparable to the complexity of of f-line algorithms for estimating a single analytic center. Finally, we brie fly discuss a number of open problems which are currently under investigati on.