Title:

Methods for Network Structure Prediction

Speaker:

Hisashi Kashima

Abstract:

Recently, there has been a surge of interest in the study of analytical methods for network structured data such as the world wide web, social networks, and biological networks. In the area of data mining and machine learning, "link mining" has become a popular subarea for analyzing such data. Among several tasks in link mining, the link prediction problem is the task of predicting unobserved portion of the network from the observed part of the network (or to predict the future structure of the network given the current structure of the network.)

Link prediction has several applications including predicting relations among participants such as friendship, and predicting their future behavior such as communications and collaborations. In the field of bioinformatics, predicting protein-protein interactions and regulatory relationships among genes can provide guidance on the design of experiments for discovering new biological facts.

This talk reviews several methods methods for the link prediction problem including; link metrics studied in the area of social network analysis and information retrieval, probabilistic relational models as general models of network structures, and supervised and semi-supervised learning methods especially well-studied in the area of bioinformatics.