Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 55249 (01)
- University: Sharif University of Technology
- Department: Industrial Engineering
- Advisor(s): Khedmati, Majid
- Abstract:
- Graphs are a language to describe and analyze connections and relations. Recent developments have increased graphs' applications in real-world problems such as social networks, researchers' collaborations, and chemical compounds. Now that we can extract graphs from real life, how can we model and generate graphs similar to a set of known graphs or that are very likely to exist but haven't been discovered yet? Therefore, this research will focus on the problem of graph generation. In graph generation, a set of graphs is a training dataset, and the goal of the thesis is to present an improved deep generative model to learn the training data's distribution, structure, and features.Identifying the distribution over the graph's characteristics is the importance of the graph generation problem. Defining a probability distribution over graphs is challenging due to graphs' complex structures with a variety of characteristics, such as degree distribution, clustering coefficient, node labels, edge labels, etc. Therefore, presenting a generative model to learn this distribution is an advantage. This research investigates the problem of graph generation using deep generative models and reviews classical models. After introducing different models in graph machine learning, we present our suggested framework to generate graphs based on graph neural networks. Then we will address the challenges in evaluating the results and propose a solution to make one of the evaluation metrics in this field more interpretable. The scope of this research includes static, undirected, and unweighted graphs; although the framework is applicable in all the cases, these properties did not study directly here.The main contribution of this thesis is offering an improved framework that competes with current models based on results in the evaluation metric of maximum mean discrepancy and the run time. Shortly, the framework learns node representations and edge probability jointly through a graph neural network and a multiple neural network.
- Keywords:
- Deep Generative Modeling ; Graph Neural Network ; Graph Representation ; Maximum Mean Discrepency ; Graph Construction
Digital Object List-
محتواي کتاب - view
Bookmark