Citation of the INFORMS Optimization Society Young Researchers Prize
Hussein Hazimeh and Rahul Mazumder are awarded the 2020 INFORMS Optimization Society Prize for Young Researchers for their paper, “Fast Best Subset Selection: Coordinate Descent and Local Combinatorial Optimization Algorithms,” Operations Research, to appear. The paper presents a novel algorithm for the Best Subset Selection problem (BSS), which is ordinary least-squares linear regression but with a penalty (or constraint) on the number of non-zero coefficients in the model, which induces sparsity. BSS is a fundamental problem in high-dimensional statistics, where sparse solutions with good explanatory power are crucial for practical use. Yet, it is NP-hard and, therefore, has been labeled computationally intractable and replaced by popular formulations such as LASSO and ridge regression. The authors combine tools and techniques from high-dimensional statistics, continuous optimization, integer programming, and open-source software development to provide the community with a highly effective heuristic for BSS, one which has strong theoretical underpinnings, outperforms existing state-of-the-art codes in many cases, and is easy to generalize to other settings.