Blind Minimax Estimation

Improving MSE over Least Squares

Welcome to the blind minimax estimation (BME) homepage. Here you can find usage examples and MATLAB implementations of the newly developed BME method. BME is a biased estimation technique for the linear regression problem, proved to dominate the commonly-used least squares estimator. In this website we describe both the simple one-shot BME and an extension for on-line systems which is designed to operate with low computation time.

This website is built as follows:

® One-shot BME introduces the blind minimax estimator, a technique dominating the least squares approach [1]. It contains an overview of the method and a Matlab example.

® On-line BME describes an application of the BME for adaptive estimation [3]. Here you will find Matlab code illustrating the improvement over the standard RLS technique.

® Links & Downloads contains BME publications, contact information, links, and a downloadable Matlab package with implementations of all estimators.††

 

"If one observes the real random variables X1...Xn independently normally distributed with unknown means ξ1...ξn and variance 1, it is customary to estimate ξi by Xi† . If the loss is the sum of squares of the errors, this estimator is admissible for n 2, but inadmissible for n ≥ 3."
C. Stein, 1956 [
5]
(also known as
Steinís phenomenon)

This website is a part of an undergraduate project
by Guy Leibovitz and Asaf Elron, under the supervision of Zvika Ben-Haim,
at the
Signal and Image Processing Lab, Faculty of Electrical Engineering,
TechnionóIsrael Institute of Technology.