Algorithmic Design for Big Data Related Optimization

Abstract

We live in the age of big data. The 5 characteristics of big data - volume, value, variety, velocity and veracity - have a significant impact on optimization. In this talk, we discuss some thinking of algorithmic design for big data related optimization problems. Specifically, we consider splitting methods for large scale structure optimization, to analyze the data with high volume and low value density. We also design efficient algorithms for distribution robust optimization, to cope with brittle veracity in data analysis. Finally, we propose LP-based approach for Markov Decision Process, which lays a deep ground in sequential decision making with dynamic data generated at a high velocity.