Home > Events > Oral Candidacy - Chuxu Zhang

Oral Candidacy - Chuxu Zhang

Start: 3/21/2019 at 10:00AM
End: 3/21/2019 at 1:00PM
Location: 384 Nieuwland
Attendees: Faculty and students are welcome to attend the presentation portion of the defense.
Add to calendar:
iCal vCal

Chuxu Zhang

Oral Candidacy Exam

March 21, 2019      10:00 am      384 Nieuwland

Adviser:  Dr. Nitesh Chawla

Committee Members:

      Dr. David Chiang      Dr. Meng Jiang      Dr. Xiangliang Zhang


Relational Representation Learning for Heterogeneous Graphs


Online web services are fast developing. Most of these real-world complex systems can be structured into heterogeneous graphs that encode rich information through multi-typed nodes, relationships, as well as contents associated with nodes. Heterogeneous graph mining has gained a lot of attention in recent years. The labor-consuming feature engineering activity, however, responds to the requirements of the heterogeneous graph mining tasks, requiring both a domain understanding and large exploratory search space for possible features. Therefore, generalizing the feature engineering activity through representation learning that automates the discovery of useful relational latent features among nodes and links poses critical challenges yet advances the research area. Moreover, a generalized feature representation from such heterogeneous graphs can lend itself to a variety of graph mining tasks (such as link prediction, node classification, etc.), mitigating the need for a purposeful feature engineering dependent on the task at hand.

In this proposal, I propose a series of relational representation learning methodologies to facilitate the generalized feature engineering and its application to a variety of tasks for validation as well as application.  The core contributions of my dissertation include: algorithms for personalized ranking in heterogeneous graphs; heterogeneous graph embedding techniques; learning to learn frameworks of relational representation in heterogeneous graphs;  and several use-case applications for graph mining. These are validated on a number of real-world data sets demonstrating the accuracy as well as generalization of the proposed work.