repository
Graph Domain Adaptation for Semi-Supervised Learning
| DC Field | Value | Language |
|---|---|---|
| dc.contributor.advisor | 신현정 | - |
| dc.contributor.author | 박성홍 | - |
| dc.date.accessioned | 2022-11-29T03:01:12Z | - |
| dc.date.available | 2022-11-29T03:01:12Z | - |
| dc.date.issued | 2022-08 | - |
| dc.identifier.other | 32236 | - |
| dc.identifier.uri | https://dspace.ajou.ac.kr/handle/2018.oak/20864 | - |
| dc.description | 학위논문(박사)--아주대학교 일반대학원 :인공지능학과,2022. 8 | - |
| dc.description.tableofcontents | 1. Introduction 1 2 Mutual Adaptation for Heterogeneous Data 5 2.1 Mutual Adaptation 6 2.1.1 Synopsis 8 2.1.2 Formulation and Optimization 12 2.1.3 Pseudo-Labeling 14 2.1.4 Mutual Label Propagation 16 2.2 Experiments 18 2.2.1 Datasets 18 2.2.2 Results for Feature Space Alignment 20 2.2.3 Performance Comparison 22 2.2.4 Ablation Study 27 3 Prospective Adaptation for Longitudinal Data 30 3.1 Prospective Adaptation 31 3.1.1 Feature Transformer 35 3.1.2 Domain Discriminator 36 3.1.3 Label Predictor 36 3.1.4 Optimization 37 3.2 Empirical Study and Evaluation 40 3.2.1 Datasets 40 3.2.2 Experimental Settings 41 3.2.3 Results for Feature Space Alignment 42 3.2.4 Performance Comparison 44 3.3 Application: Alzheimer’s Disease Conversion Prediction 45 3.3.1 Background of Alzheimer’s Disease 45 3.3.2 Formulation: Brain Transition 49 3.3.3 Formulation: Conversion Risk 50 3.3.4 Optimization 51 3.3.5 Experimental Results 54 3.3.6 Enrichment Study 59 4 Multiplex Adaptation for Multi-Modal Data 63 4.1 Multiplex Adaptation 64 4.1.1 Background and Synopsis 67 4.1.2 Formulation 70 4.1.3 Optimization 71 4.1.4 Label Prediction 74 4.2 Experiments 75 4.2.1 Artificial Data – Empirical Study 75 4.2.2 Benchmark Data – Performance Comparison 77 4.3 Application: Historical Faction Identification 79 4.3.1 Background and Data 81 4.3.2 Experimental Settings – Network Construction 83 4.3.3 Results – Performance Comparison 84 5 Conclusion 86 References 89 | - |
| dc.language.iso | eng | - |
| dc.publisher | The Graduate School, Ajou University | - |
| dc.rights | 아주대학교 논문은 저작권에 의해 보호받습니다. | - |
| dc.title | Graph Domain Adaptation for Semi-Supervised Learning | - |
| dc.type | Thesis | - |
| dc.contributor.affiliation | 아주대학교 일반대학원 | - |
| dc.contributor.department | 일반대학원 인공지능학과 | - |
| dc.date.awarded | 2022. 8 | - |
| dc.description.degree | Doctoral | - |
| dc.identifier.localId | T000000032236 | - |
| dc.identifier.uci | I804:41038-000000032236 | - |
| dc.identifier.url | https://dcoll.ajou.ac.kr/dcollection/common/orgView/000000032236 | - |
| dc.subject.keyword | Domain Adaptation | - |
| dc.subject.keyword | Graph | - |
| dc.subject.keyword | Semi-Supervised Learning | - |
| dc.description.alternativeAbstract | Graphs are used in wide range of domains, such as bioinformatics, text analytics, and social network in that graphs can capture associations of various sources of data while managing large amount of data. In real-world, graphs exist in the form of multiple data. For example, there are longitudinal data generated according to the passage of time, multi-modal data including various relational information, or heterogeneous data from different domains. Therefore, when applying the machine learning algorithm to graphs, learning multiple data together rather than single data has the advantage of utilizing more information. However, learning multiple data in a simple way can be rather detrimental due to the different properties of data and domains. For this reason, a process of transforming multiple data to be regarded as one set is required, and this process should be subdivided according to data types. In this dissertation, a domain adaptation-based framework for semi-supervised learning is proposed, which can learn multiple graph data in an integrated way. The proposed framework divides data types into three cases for efficient learning and includes methods suitable for each: mutual adaptation for heterogeneous data, prospective adaptation for longitudinal data, and multiplex adaptation for multi-modal data. Each method enables comprehensive data integration and learning, and some real-world applications such as Alzheimer’s disease conversion prediction and historical faction identification can also be successfully performed. | - |
- Appears in Collections:
- Graduate School of Ajou University > Department of Artificial Intelligence > 4. Theses(Ph.D)
- Files in This Item:
- There are no files associated with this item.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.
-
- License
- STATISTICS
- Total Visit :5,302,466
- Total Download :2,140
- Today View :3,104
