Domain generalization method based on cross-space multi-scale information aggregation and inference consistency
In machine learning, it typically assumes that training data and testing data of models are drawn from the same distribution. However, in real-world applications, data distributions often differ, resulting in domain shift problems that adversely affect model generalization. Existing domain generaliz...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
POSTS&TELECOM PRESS Co., LTD
2025-06-01
|
| Series: | 智能科学与技术学报 |
| Subjects: | |
| Online Access: | http://www.cjist.com.cn/zh/article/doi/10.11959/j.issn.2096-6652.202516/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In machine learning, it typically assumes that training data and testing data of models are drawn from the same distribution. However, in real-world applications, data distributions often differ, resulting in domain shift problems that adversely affect model generalization. Existing domain generalization methods primarily focus on extracting domain-invariant features while overlooking the potential impact of domain-specific features on model predictions. To address this issue, a domain discriminator based on cross-space multi-scale information aggregation was proposed. By capturing multi-scale information, domain-specific features were effectively removed and the extraction of domain-invariant features was enhanced. Additionally, the momentum update inference consistency loss function was employed to leverage the inference consistency of sample category centers, further improving model robustness. Comparative experiments and analysis conducted on multiple public datasets demonstrate that the proposed method exhibits superior performance in domain generalization, effectively mitigating the impact of domain-specific features on model performance and providing a technical reference for addressing domain shift problems. |
|---|---|
| ISSN: | 2096-6652 |