TRIPODS RWG 5

Extracting information from data and making robust decisions requires effective modeling techniques and efficient solution algorithms. Mathematical optimization models play a major role for data analytics, such as regression, classification, clustering, and dimension reduction through regularization or sparse optimization. Though many optimization models and algorithms have been developed in past several decades, there are many new situations faced in the big data era. Besides the volume of the data, variety, velocity, veracity and variability issues make many classic machine learning models either ineffectiveness, or put computational burden on obtaining solutions. For example, the uncertainty of data in veracity may cause most of the models intractable, and the changes of the input (velocity) require online decisions through super fast algorithms. Additionally, the non-smooth and non-convex characteristics of many models need the development of efficient algorithms.

To tackle these challenges, advanced robust and high-confidence models and scalable algorithms are needed, as well as novel methodologies and powerful computing technologies such as parallel computing techniques and distributed computing systems. The team will utilize its interdisciplinary expertise in statistics, mathematics, and optimization to construct data-driven optimization models and intelligent algorithms through the consideration of data characteristics from the bottom.