Theme：Handling Class Imbalance and Small Sample Issues: Foundation, Algorithms, and Applications

Reporter：Professor Zidong WANG, Brunel University

Time：09:30-11:00,  26th June, 2025

Venue：Building 2, room 2072, iHarbour

Abstract：

In big data analysis, it is usually difficult to collect high-quality labels, and this leads to two issues in deep learning, namely, the class imbalance issue and the small sample issue. In this talk, we first introduce some background knowledge about the deep learning from the perspectives of concepts, techniques, applications and challenges. Then, we introduce three state-of-the-art algorithms for solving the class imbalance and small sample issues: 1) a novel contrastive adversarial network for minor-class data augmentation; 2) a novel subdomain-alignment data augmentation approach; and 3) a novel prototype-assisted contrastive adversarial network for weak-shot learning. All the three algorithms are applied to pipeline fault diagnosis, which outperform existing ones. Finally, we conclude our main contributions and some future directions.