Wire · technology
A novel disentangled framework to tackle class imbalance in semi-supervised learning
◆ Sectors
◆ Geography
◆ Source
◆ Verified
Fusion42 · 5 November 2026 · Fusion42 review
Researchers from Southeast University introduced DALE, a novel framework that disentangles pseudo-label generation from classification to tackle class imbalance in semi-supervised learning, significantly improving performance on skewed datasets like CIFAR-10-LT and Small-ImageNet127.
This Wire brief sits within Fusion42's coverage of AI Infrastructure and MLOps.
◆ ◆ The Wire takeaway
You can shift your semi-supervised learning models with DALE’s approach to separate pseudo-label generation from classification, reducing bias on imbalanced data and boosting accuracy. Test this framework on your datasets to overcome skew-related performance drops now.
◆ Coverage
1 source · 5 Nov 2026
◆ Related on Wire
◆ Topics