Presented by Janet Layne, Computer Science emphasis
Graph Representation Learning (GRL) methods which effectively capture a node’s neighborhood structure in their representations can show excellent performance on important machine learning tasks such as node and graph classification. This project aims to enhance the state-of-the-art in structural graph representation learning via three different avenues. The first goal is to enable transfer learning for our existing high-performing structural GRL method via algorithm modification. The second aims to develop a temporal structural GRL approach for use with dynamic graphs. Existing temporal methods are limited in terms of efficiency and effectiveness: they scale poorly to even moderate number of timestamps, or capture structural role only tangentially. Thus, there is great need for an efficient structural temporal GRL approach. The third objective is to create a custom attribution procedure each for static and temporal GRL approaches to identify the graph components important to the outcome of classification tasks. A portion of the proposed work is complete and shows promising experimental results. Completed work will be summarized, and plans for fulfillment of each of the three defined objectives will be described.
Dr. Edoardo Serra (Advisor) , Dr. Francesca Spezzano, Dr. Sole Pera, Dr. Marion Scheepers