Efficient architectures to recover the regularized least squares solution

Several practical applications are concerned with the identification of the least squares (LS) solution. The objective is to attain this solution accurately and efficiently while conserving resources. The computational and storage requirements to determine the LS solution by any iterative procedure become prohibitively large as the problem dimensions grow. This work presents some architectures based on thresholded binary networks which recover regularized LS solutions by partitioning such networks and adopting a switching operation between active and inactive partitions to optimize the objective function. Also, an iterative method based on steepest descent is briefly discussed and implemented. It yields reliable estimates of the regularized LS solution, while providing savings in computation and storage.