Title Page
Contents
Abstract 7
Chapter 1. Introduction 8
Chapter 2. Related Work 10
Chapter 3. Adversarial Adaptation with Distillation 13
3.1. Step 1: Fine-tune the source encoder and the classifier 14
3.2. Step 2: Adapt the target encoder via adversarial adaptation with distillation 14
3.3. Step 3: Test the target encoder on the target data 16
Chapter 4. Experiments 17
Chapter 5. Conclusion and Future Work 22
References 23
Appendix 28
A. Hyperparameters and Training Details 28
B. Hyperparameters and Training Details 29
국문 요지 31
Table 1. Sentiment classification accuracy with BERT BASE uncased model on 30 cross-domain... 18
Table 2. Experimental results with varying temperature values. * denotes that the value is greater... 20
Table 3. Sentiment classification accuracy with distil BERT BASE uncased model on 30 cross-... 29
Table 4. Sentiment classification accuracy with RoBERTa BASE model on 30 cross-domain... 30
Figure 1. An overview of Adversarial Adaptation with Distillation. In Step 1, we first fine-tune a... 13