L SVRG

L-SVRG (Loopless Stochastic Variance Reduced Gradient) is a stochastic optimization algorithm designed to efficiently minimize large-scale objective functions, particularly those arising in machine learning. Current research focuses on extending L-SVRG to Riemannian manifolds, decentralized settings, and applications involving non-convex objectives, often incorporating techniques like acceleration, gradient tracking, and adaptive sampling to improve convergence rates and scalability. These advancements are significant because they enable the application of efficient optimization methods to increasingly complex problems in scientific machine learning and other fields requiring large-scale data analysis, such as recommendation systems and IoT data completion.