New Deep Neural Networks for Unsupervised Feature Learning on Graph Data

Gao, Hongchang (2020) New Deep Neural Networks for Unsupervised Feature Learning on Graph Data. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

Preview

PDF Updated Version Download (3MB) | Preview

Abstract

Graph data are ubiquitous in the real world, such as social networks, biological networks. To analyze graph data, a fundamental task is to learn node features to benefit downstream tasks, such as node classification, community detection. Inspired by the powerful feature learning capability of deep neural networks on various tasks, it is important and necessary to explore deep neural networks for feature learning on graphs. Different from the regular image and sequence data, graph data encode the complicated relational information between different nodes, which challenges the classical deep neural networks. Moreover, in real-world applications, the label of nodes in graph data is usually not available, which makes the feature learning on graphs more difficult.
To address these challenging issues, this thesis is focusing on designing new deep neural networks to effectively explore the relational information for unsupervised feature learning on graph data.

First, to address the sparseness issue of the relational information, I propose a new proximity generative adversarial network which can discover the underlying relational information for learning better node representations. Meanwhile, a new self-paced network embedding method is designed to address the unbalance issue of the relational information when learning node representations. Additionally, to deal with rich attributes associated to nodes, I develop a new deep neural network to capture various relational information in both topological structure and node attributes for enhancing network embedding. Furthermore, to preserve the relational information in the hidden layers of deep neural networks, I develop a novel graph convolutional neural network (GCN) based on conditional random fields, which is the first algorithm applying this kind of graphical models to graph neural networks in an unsupervised manner.

---

Share

Citation/Export:	Select format... Citation - Text Citation - HTML Endnote BibTex Dublin Core OpenURL MARC (ISO 2709) METS MODS EP3 XML Reference Manager Refer
Social Networking:	Share |

---

Details

Item Type:	University of Pittsburgh ETD
Status:	Unpublished
Creators/Authors:	Creators Email Pitt Username ORCID Gao, Hongchang hog10@pitt.edu hog10
ETD Committee:	Title Member Email Address Pitt Username ORCID Committee Chair Huang, Heng heng.huang@pitt.edu Committee Member Zhi-Hong, Mao Committee Member Wei, Gao Committee Member Liang, Zhan Committee Member Wei, Chen
Date:	28 September 2020
Date Type:	Publication
Defense Date:	19 June 2020
Approval Date:	28 September 2020
Submission Date:	15 July 2020
Access Restriction:	No restriction; Release the ETD for access worldwide immediately.
Number of Pages:	120
Institution:	University of Pittsburgh
Schools and Programs:	Swanson School of Engineering > Electrical and Computer Engineering
Degree:	PhD - Doctor of Philosophy
Thesis Type:	Doctoral Dissertation
Refereed:	Yes
Uncontrolled Keywords:	unsupervised feature learning, deep neural networks, graphs
Date Deposited:	28 Sep 2020 19:21
Last Modified:	28 Sep 2020 19:21
URI:	http://d-scholarship.pitt.edu/id/eprint/39284

---

Metrics

Monthly Views for the past 3 years

Plum Analytics

---

Actions (login required)

View Item