Community Detection: Instead of optimizing some user-defined criteria, we consider community detection from a new point of view: local distance dynamics.The basic idea is to envision a network as a dynamical system, and each agent interacts with its local partners. Instead of investigating the node dynamics, we actually examine the change of “distances” among linked nodes. As time evolves, these distances will be shrunk or stretched gradually based on their topological structures. Finally all distances among linked nodes will converge into a stable pattern, and communities can be intuitively identified. [see http://arxiv.org/abs/1409.7978]
LocalSVM: This project is mainly about the improving of SVM. We parttion the whole data set into the global data which will be classified by libSVM and local data which will be classified by KNN or other classfication method with good local generalization ability. In this way ,we hope to improve the local generalization ability of the SVM with the global generalization ability hold.
Distributed Data Stream Classification: Our project focuses on how to learn the association among concept drift data streams, aiming to combine information from local streams to get better prediction. The project is divided into local learning part and global learning part. From local part, we build P-Tree to maintain import data for each stream and use Error-driven Representativeness Learning to update P-tree due based on weight vector of data. We also use PCA and Statistical Analysis to capture abrupt concept-drifting. From global part, we using Weighted Majority to get final prediction. This learning method provides a new learning framework based on typical data which reflect streams association.