Graph embedding algorithms for attributed and temporal graphs

Abstract

Palash Goyal is a Senior Research Scientist at Samsung Research America. In 2019, he got his doctoral degree in Computer Science from University of Southern California under the advisory of Dr. Emilio Ferrara. Over the course of his PhD, he worked with several government funded projects including IARPA and DARPA spanning domains of time series prediction, text analysis and understanding social behavior. He also applied his work on graph embedding in various industrial settings including developing automation systems in Siemens Corporate, identifying key steam injection candidate oil wells for Chevron and developing recommendation system for Target. His work on tracking temporal evolution of graphs using non-timestamped data was nominated for best paper award in ACM Hypertext. In his thesis, supervised by Dr. Emilio Ferrara (USC Information Sciences Institute), he ex- tended the work of learning low-dimensional representations of nodes in a graph in several di- rections. Firstly, he published a survey of existing graph embedding approaches and shed light into the dependency between them drawing insights into the relations effectively captured by the methods. He further proposed a benchmark to evaluate any graph embedding approach and un- derstand its applicability. Secondly, he built models to capture temporal patterns in sequential graphs and efficiently update embeddings for streaming graphs. Finally, he developed multi-modal models for learning representations through graph data as well as other forms of data available for the nodes and edges including text.