Aug 05, 2020
- ZOOM Cloud Meetings
- Metting ID:8572571104
Similar to their deterministic counterpart, stochastic second order methods are aimed at accelerating, robustifying, and enhancing the performance of first order algorithms via incorporating stochastic or approximate curvature information in the algorithmic design. Albeit the general popularity and high prevalence of first order schemes, stochastic second order-type methods have recently gained increasing attention and have been successfully applied to solve challenging large-scale learning tasks, reinforcement learning problems, and other big data applications.
In this talk, we present novel extensions of these second order approaches to a class of structured nonsmooth and nonconvex optimization problems. The proposed methods utilize stochastic or deterministic higher order information to generate semismooth Newton or quasi-Newton steps for an underlying prox-type fixed-point equation that represents the first order optimality conditions of the problem. Besides the algorithmic ideas and foundations, we will also discuss convergence results and variants using variance reduction techniques that enjoy better complexity bounds. Numerical comparisons on large-scale logistic regression, sparse deep learning, and other problems are provided illustrating the efficiency of nonsmooth second order-type methods.
AIRS Principal Investigator, Assistant Professor at school of data science at CUHKSZ
Andre Milzarek is an assistant professor at school of data science (SDS) at the Chinese University of Hong Kong, Shenzhen. He received his master’s degree with honours and his doctoral degree in mathematics from the Technical University of Munich in Germany under the supervision of Michael Ulbrich in 2013 and 2016. He was a postdoctoral researcher at the Beijing International Center for Mathematical Research at the Peking University in Beijing from 2017 to March 2019. He is a member of the Shenzhen research institute of big data (SRIBD) and the Shenzhen institute of artificial intelligence and robotics for society (AIRS). His main research directions and interests cover nonsmooth optimization, large-scale and stochastic optimization, second order methods and theory. From 2010 to 2012 he was supported by the Max-Weber program of the state of Bavaria and in 2017 he received the Boya Postdoctoral Fellowship at Peking University.