Title Page
ABSTRACT
초록
Contents
NOMENCLATURE 14
CHAPTER 1. INTRODUCTION 16
CHAPTER 2. RELATED WORK 20
CHAPTER 3. PROPOSED METHOD 22
3.1. Encoding and Decoding Process 24
3.2. Type-specific Conditioner 26
3.3. Generator 29
3.4. Discriminator 31
3.5. Loss 32
CHAPTER 4. EXPERIMENTS 33
4.1. Evaluation Metric 34
4.2. Evaluation Results 35
4.2.1. Results of Statistical Similarity 35
4.2.2. Results of Machine Learning Efficiency 39
4.2.3. Results of Privacy Preservability 41
4.3. Ablation Study 44
CHAPTER 5. CONCLUSION 46
REFERENCES 47
Table 1. Detailed information about six datasets 33
Table 2. Statistical differences for six datasets 38
Table 3. MLE results for classification datasets 40
Table 4. MLE results for regression datasets 41
Table 5. NNDR results for six datasets 43
Table 6. Ablation study on classification datasets 44
Table 7. Ablation study on regression datasets 45
Figure 1. Tabular data synthesis using interpolative CDF decoding and type- specific conditioner. Original data are converted by encoding, and then discrete and... 23
Figure 2. Example of generating a condition cond using a type-specific conditioner. Yellow is a discrete condition, blue is a condition for a binned... 28
Figure 3. Comparison of correlations between variables with the proposed and comparison models on the Shoppers. (a) and (b) are correlation plots of the original... 36
Figure 4. DCR plot for California House 42