Chao-Hao Hsu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Chao-Hao Hsu has a total of 2 co-authors. Here are the top co-authors by number of shared publications:
- Hsin-Chang Tsai
Institute of Communications Engineering, National Yang Ming Chiao Tung University, Hsinchu, Taiwan
1 shared publication - Ming-Chun Lee
Institute of Communications Engineering, National Yang Ming Chiao Tung University, Hsinchu, Taiwan
1 shared publication
Latest Publications
- Open Access
- 3 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 23508-23525
Fault and Severity Diagnosis Using Deep Learning for Self-Organizing Networks With Imbalanced and Small Datasets
With the growing complexity of wireless networks, manual management of networks becomes infeasible. To address this, self-organizing networks (SONs) have been introduced to provide solutions by offering self-organizing approaches to networks. Developing effective self-organizing approaches often depends on data-driven or learning-based methods, which require well-structured and balanced datasets. However, in practical scenarios, datasets are often imbalanced or even very small. To address this issue from the fault diagnosis aspect of SONs, this paper investigates the learning-based fault and severity diagnosis approaches under imbalanced and small datasets for wireless networks. We first propose a deep learning-based diagnosis framework, in which the diagnosis problem can be cast as a regression problem. Then, several approaches, including re-weighting, distribution smoothing, and balanced MSE, that can be used to resolve the imbalanced issue for regression problem are examined under the diagnosis purpose. Subsequently, to resolve the issue that the amount of data samples for diagnosis could be few, model pre-training and meta-learning-based approaches are used, aiming to quickly adapt the pre-trained diagnosis model to the targeting scenarios for diagnosis. Extensive simulation results based on realistic setups are conducted to evaluate our proposed approaches. Results show that our approaches can effectively diagnose the faults and their severity and outperform the baseline approaches under imbalanced and small datasets.
- 3 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 23508-23525
Open AccessFault and Severity Diagnosis Using Deep Learning for Self-Organizing Networks With Imbalanced and Small Datasets
With the growing complexity of wireless networks, manual management of networks becomes infeasible. To address this, self-organizing networks (SONs) have been introduced to provide solutions by offering self-organizing approaches to networks. Developing effective self-organizing approaches often depends on data-driven or learning-based methods, which require well-structured and balanced datasets. However, in practical scenarios, datasets are often imbalanced or even very small. To address this issue from the fault diagnosis aspect of SONs, this paper investigates the learning-based fault and severity diagnosis approaches under imbalanced and small datasets for wireless networks. We first propose a deep learning-based diagnosis framework, in which the diagnosis problem can be cast as a regression problem. Then, several approaches, including re-weighting, distribution smoothing, and balanced MSE, that can be used to resolve the imbalanced issue for regression problem are examined under the diagnosis purpose. Subsequently, to resolve the issue that the amount of data samples for diagnosis could be few, model pre-training and meta-learning-based approaches are used, aiming to quickly adapt the pre-trained diagnosis model to the targeting scenarios for diagnosis. Extensive simulation results based on realistic setups are conducted to evaluate our proposed approaches. Results show that our approaches can effectively diagnose the faults and their severity and outperform the baseline approaches under imbalanced and small datasets.