Exploring Shifted Domain Problem within MNIST Dataset
Authors:
- 16 February 2023
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 750-754
Abstract
Domain data can be shifted in any direction so it will be shared in different distributions to its original domain. This could be a problem since the model was trained with different distributions. It is found that adversarial domain adaptation using domain adversarial neural networks (DANN) can help to solve this problem on some scale. DANN can minimize the discrepancy between source and target data so the model can work well in both domains. The experiment is done by utilizing MNIST dataset that shifted into some conditions. In a condition when the shifting of distribution is too far, DANN is struggling to maintain the knowledge extracted from source data which leads to underperformance in the source and target domain. In contrast, when the shifting is closer, DANN can easily fit the model so it can perform well in both domains. It proves DANN is one of the good approaches to performing domain adaptation in small discrepancies.Keywords
This publication has 17 references indexed in Scilit:
- Contrastive Adaptation Network for Single- and Multi-Source Domain AdaptationIEEE Transactions on Pattern Analysis and Machine Intelligence, 2020
- Adversarial Network With Multiple Classifiers for Open Set Domain AdaptationIEEE Transactions on Multimedia, 2020
- A Survey of Unsupervised Deep Domain AdaptationACM Transactions on Intelligent Systems and Technology, 2020
- Learning Target-Domain-Specific Classifier for Partial Domain AdaptationIEEE Transactions on Neural Networks and Learning Systems, 2020
- Discriminative Adversarial Domain AdaptationProceedings of the AAAI Conference on Artificial Intelligence, 2020
- Rethinking Triplet Loss for Domain AdaptationIEEE Transactions on Circuits and Systems for Video Technology, 2020
- A Survey of Domain Adaptation for Machine TranslationJournal of Information Processing, 2020
- Self-Supervised Domain Adaptation for Computer Vision TasksIEEE Access, 2019
- Confidence Measures for Deep Learning in Domain AdaptationApplied Sciences, 2019
- Beyond Sharing Weights for Deep Domain AdaptationIEEE Transactions on Pattern Analysis and Machine Intelligence, 2018