We propose a novel technique that exploits some interesting properties of the Beta distribution to derive a sparse solution to the traditional general linear regression under the Gaussian noise assumption. Our proposed technique provides a theoretically, conceptually and computationally better alternative to both the LASSO and the relevance vector machine in the sense that it is centered around an objective function that is convex and easy to interpret. We demonstrate the strength of our proposed technique through examples, and we also provide a theoretical proof of the merits of our method.
Department, Program, or Center
The John D. Hromi Center for Quality and Applied Statistics (KGCOE)
Fokoue, Ernest, "Beta Induced Sparsity Algorithm" (2000). Accessed from
RIT – Main Campus
Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works in February 2014.