Events Calendar

PhD Defense - Palash Goyal

PhD Candidate: Palash Goyal
Date:
Tuesday July 23rd, 2019
1:00 p.m. - 3:00 pm
Location: OHE 136
Committee: Emilio Ferrara, Gaurav Sukhatme, Cauligi Raghavendra

Title:
Graph Embedding Algorithms for Attributed and Temporal Graphs

Abstract:
Learning low-dimensional representations of nodes in a graph has recently gained attention due to its wide applicability in network tasks such as graph visualization, link prediction, node classification and clustering. The models proposed often preserve certain characteristic properties of the graph and are tested on these tasks. In this thesis, I propose to extend the graph embedding work in three directions. Firstly, I yield insights into the existing models and understand the universality of the learned embeddings and embedding methods. Specifically, I study the dependence of performance on a task on the model hyperparameters of the embedding method. I also analyze the characteristics of models required for each network task. Further, I propose a benchmark to evaluate any graph embedding approach and draw insights into it. Secondly, I propose an extension of graph embedding approach which can capture edge attributes of a graph. I show that capturing such attributes can be useful in link prediction and propose a model to learn edge attributes along with higher order proximity and social roles. Thirdly, I build models which can update embeddings efficiently for streaming graphs and can capture temporal patterns in sequential graphs.

Location: Olin Hall of Engineering (OHE) - 136

Audiences: Everyone Is Invited

Contact: Lizsl De Leon