Unbiased Least Squares Regression via Averaged Stochastic Gradient Descent

Published Online:https://doi.org/10.1287/moor.2024.0660

We consider an online least squares regression problem with optimal solution θ* and Hessian matrix H, and study a time-average stochastic gradient descent estimator of θ*. For k2, we provide an unbiased estimator of θ* that is a modification of the time-average estimator and runs with an expected number of time-steps of order k, with O(1/k) expected excess risk. The constant behind the O notation depends on parameters of the regression and is a polylogarithmic function of the smallest eigenvalue of H. We provide both a biased and unbiased estimator of the expected excess risk of the time-average estimator and of its unbiased counterpart, without requiring knowledge of either H or θ*. We describe an “average-start” version of our estimators with similar properties. Our approach is based on randomized multilevel Monte Carlo methods. Numerical experiments confirm our theoretical findings.

Supplemental Material: The online appendix is available at https://doi.org/10.1287/moor.2024.0660.

INFORMS site uses cookies to store information on your computer. Some are essential to make our site work; Others help us improve the user experience. By using this site, you consent to the placement of these cookies. Please read our Privacy Statement to learn more.