Comparative Analysis on Classical Meta-Metric Models for Few-Shot Learning
Few-shot learning are methods and scenarios learned from a small amount of labeled data. While recent meta-metric learning methods have made significant progress, there are still questions about what is the key point of these methods and how they work. To address these problems, in this paper, we ev...
Main Authors: | , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
IEEE
2020-01-01
|
Series: | IEEE Access |
Subjects: | |
Online Access: | https://ieeexplore.ieee.org/document/9139379/ |
id |
doaj-ca5752686d0343aea9731217e25afccf |
---|---|
record_format |
Article |
spelling |
doaj-ca5752686d0343aea9731217e25afccf2021-03-30T02:08:41ZengIEEEIEEE Access2169-35362020-01-01812706512707310.1109/ACCESS.2020.30086849139379Comparative Analysis on Classical Meta-Metric Models for Few-Shot LearningSai Yang0https://orcid.org/0000-0002-3256-7306Fan Liu1https://orcid.org/0000-0001-8746-9845Ning Dong2https://orcid.org/0000-0003-3045-9798Jiaying Wu3https://orcid.org/0000-0001-5184-4295School of Electrical Engineering, Nantong University, Nantong, ChinaCollege of Computer and Information, Hohai University, Nanjing, ChinaGraduate School of ISEE, Kyushu University, Fukuoka, JapanGraduate School of IPS, Waseda University, Fukuoka, JapanFew-shot learning are methods and scenarios learned from a small amount of labeled data. While recent meta-metric learning methods have made significant progress, there are still questions about what is the key point of these methods and how they work. To address these problems, in this paper, we evaluate the effects of different parts in classical models. To be specific, we 1) use four typical networks AlexNet, VGG16, GoogLeNet, and ResNet50 to replace the original feature extraction part in the Matching Network, Prototypical Network, and Relation Network, and compare the best results with 17 state-of-the-art meta-metric learning algorithms. 2) fix the feature extraction part of the Matching Network, Prototypical Network and Relation Network, and change the similarity measurement part of each into L1, L2, Cosine. 3) conduct above three models on datasets of different granularity. The experimental results show that for all models evaluated, the addition of non-pretrained networks will make the classification results worse, which shows that it is easy to overfit when using deep networks for few-shot learning. Changes in similarity measurement methods have a significant impact on results, which shows the importance to choose a suitable measurement. Moreover, there are differences in performance on different granularity datasets.https://ieeexplore.ieee.org/document/9139379/Few-shotmeta-metric learningfeature extractionsimilarity measurementcomparative analysis |
collection |
DOAJ |
language |
English |
format |
Article |
sources |
DOAJ |
author |
Sai Yang Fan Liu Ning Dong Jiaying Wu |
spellingShingle |
Sai Yang Fan Liu Ning Dong Jiaying Wu Comparative Analysis on Classical Meta-Metric Models for Few-Shot Learning IEEE Access Few-shot meta-metric learning feature extraction similarity measurement comparative analysis |
author_facet |
Sai Yang Fan Liu Ning Dong Jiaying Wu |
author_sort |
Sai Yang |
title |
Comparative Analysis on Classical Meta-Metric Models for Few-Shot Learning |
title_short |
Comparative Analysis on Classical Meta-Metric Models for Few-Shot Learning |
title_full |
Comparative Analysis on Classical Meta-Metric Models for Few-Shot Learning |
title_fullStr |
Comparative Analysis on Classical Meta-Metric Models for Few-Shot Learning |
title_full_unstemmed |
Comparative Analysis on Classical Meta-Metric Models for Few-Shot Learning |
title_sort |
comparative analysis on classical meta-metric models for few-shot learning |
publisher |
IEEE |
series |
IEEE Access |
issn |
2169-3536 |
publishDate |
2020-01-01 |
description |
Few-shot learning are methods and scenarios learned from a small amount of labeled data. While recent meta-metric learning methods have made significant progress, there are still questions about what is the key point of these methods and how they work. To address these problems, in this paper, we evaluate the effects of different parts in classical models. To be specific, we 1) use four typical networks AlexNet, VGG16, GoogLeNet, and ResNet50 to replace the original feature extraction part in the Matching Network, Prototypical Network, and Relation Network, and compare the best results with 17 state-of-the-art meta-metric learning algorithms. 2) fix the feature extraction part of the Matching Network, Prototypical Network and Relation Network, and change the similarity measurement part of each into L1, L2, Cosine. 3) conduct above three models on datasets of different granularity. The experimental results show that for all models evaluated, the addition of non-pretrained networks will make the classification results worse, which shows that it is easy to overfit when using deep networks for few-shot learning. Changes in similarity measurement methods have a significant impact on results, which shows the importance to choose a suitable measurement. Moreover, there are differences in performance on different granularity datasets. |
topic |
Few-shot meta-metric learning feature extraction similarity measurement comparative analysis |
url |
https://ieeexplore.ieee.org/document/9139379/ |
work_keys_str_mv |
AT saiyang comparativeanalysisonclassicalmetametricmodelsforfewshotlearning AT fanliu comparativeanalysisonclassicalmetametricmodelsforfewshotlearning AT ningdong comparativeanalysisonclassicalmetametricmodelsforfewshotlearning AT jiayingwu comparativeanalysisonclassicalmetametricmodelsforfewshotlearning |
_version_ |
1724185783013736448 |