Domain adaption for few-shot learning in image classification
Machine learning has been widely used in various fields and successfully addresses many image classification problems in the presence of sufficient samples, but performs poorly in the absence of samples. Few-shot learning is an innovative approach to solving this problem. In this article, I con...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis-Master by Coursework |
| Language: | English |
| Published: |
Nanyang Technological University
2023
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/164914 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Machine learning has been widely used in various fields and successfully addresses
many image classification problems in the presence of sufficient samples, but performs
poorly in the absence of samples. Few-shot learning is an innovative approach to
solving this problem. In this article, I conduct a comprehensive survey and categorize
few-shot learning algorithms from three perspectives: (i) initialization based methods;
(ii) distance metric learning based methods; and (iii) hallucination based methods. I
further study and evaluate (1) Fine-tuning; (2) Model-Agnostic Meta-Learning; (3)
Prototypical Networks; (4) Relation Networks using the Omniglot and miniImageNet
datasets. |
|---|