Yannick Baraud - UNS, Laboratoire J.A. Dieudonné, Nice. - Keywords: Statistical estimation, robust estimation

Yannick Baraud - UNS, Laboratoire J.A. Dieudonné, Nice. - Keywords: Statistical estimation, robust estimation

Contribution title: Robust estimation in statistic

The aim of mathematical statistic is to infer pieces of information on a random phenomenon one is studying from the observation of the data it produces. The problem of statistical estimation is to provide approximative values for the parameters involved in a model that is designed to describe this phenomemon. A common drawback of most of the statistical procedures (maximum likelihood, minimum contrast,...) lies in their lack of robustness which means that the estimation of these parameters can be quite good as long as the model exactely describes reality but may be terrible when the model is only approximate, which is the common situation in practice. The aim of this talk is to present a new estimation procedure that is proved to possess the property to result in robust estimators of the parameters, which means that they still perform well even when the model is not exact but only provides a good approximation of reality.