Abstract

Proceedings Abstracts of the Twenty-Fourth International Joint Conference on Artificial Intelligence

Convergence of Common Proximal Methods for L1-Regularized Least Squares / 3849
Shaozhe Tao, Daniel Boley, Shuzhong Zhang
PDF

We compare the convergence behavior of ADMM (alternating direction method of multipliers), [F]ISTA ([fast] iterative shrinkage and thresholding algorithm) and CD (coordinate descent) methods on the model L1-regularized least squares problem (aka LASSO). We use an eigen analysis of the operators to compare their local convergence rates when close to the solution. We find that, when applicable, CD is often much faster than the other iterations, when close enough to the solution. When far from the solution, the spectral analysis implies that one can often get a sequence of iterates that appears to stagnate, but is actually taking small constant steps toward the solution. We also illustrate how the unaccelerated ISTA algorithm can sometimes be faster compared to FISTA when close enough to the solution.