"Revisit Stochastic Optimization Methods in Modern Machine Learning"Tianyu WangData-driven machine learning is now widely applied across numerous fields. In this discussion, we explore key challenges in stochastic optimization methods for modern machine learning, and propose potential solutions. Specifically, we 1. examine convergence analysis in terms of stopping times and 2. demonstrate how to leverage intrinsic low-rank structures in finite-sum optimization problems. |
« back