Large-scale network : a scalable learning algorithm and visualization

Abstract

The amount of available data is predicted to be more than thousands of gigabytes per human by 2020, and current technologies are connecting people together. Data on observed actions become more available which are able to give insights on the underlying connections between individuals. However, the growing size of data presents challenges for existing machine learning methods and visualization platform. In this thesis, I focus on two problems. First, I extend an existing network learning method to large-scale networks with alternating direction method of multipliers. Testing the method with synthetic datasets, I show that the algorithm achieves similar performance with less computation time. Second, I build a tool for large-scale network data exploration. The tool is tested on several large-scale real-world datasets to illustrate its benefits.