Title Page
Abstract
국문요약
Contents
Chapter 1. Introduction 13
Chapter 2. Related Works 16
2.1. Neural Architecture Search with Evolutionary Algorithm 16
2.2. Graph Neural Architecture Search 17
2.3. Semi-Supervised Learning 18
Chapter 3. Graph Neural Architecture Search 20
3.1. Problem Statement 20
3.2. Search Space 21
Chapter 4. Proposed Algorithm 25
4.1. Overall Algorithm 25
4.2. Population Initialization 26
4.3. Offspring Generation 27
4.4. Parameter Sharing 30
Chapter 5. Experiments 33
5.1. Datasets 33
5.2. Baseline Selection 34
5.3. Experimental Settings 36
5.4. Results 36
Chapter 6. Conclusion 41
References 42
Table 3.1. The structure search space for GNN layers 22
Table 3.2. The hyperparameter search space for GNN layers 24
Table 5.1. Descriptions of graph dataset 33
Table 5.2. Test accuracy (%) comparison of semi-supervised learning in the node classification task. 37
Table 5.3. Graph neural architecture search time 38
Table 5.4. Comparison of GNN NAS methods in terms of computational costs 39
Figure 4.1. Crossover 28
Figure 4.2. Mutation 29
Figure 4.3. Parameter sharing from a single parent 31
Figure 4.4. Parameter sharing from two parents 31
Figure 4.5. Parameter sharing influenced by mutation 32
Algorithm 1. EGNAS 25
Algorithm 2. Offspring Generation 27