Citation: Thuy Thuy Le, Nhut Minh Nguyen, Nhat Truong Pham, Phuong-Nam Tran, Nguyen Doan Hieu Nguyen, Phuong Luu Vo, Balachandran Manavalan, and Duc Ngoc Minh Dang, “Federated Semi-Supervised FixMatch: Enhancing CutMix for Medical Image Segmentation”, IEEE BigData 2025 - rank B, Macau, China.
We are excited to share our new work, “Federated Semi-Supervised FixMatch: Enhancing CutMix for Medical Image Segmentation,” conducted by AiTA Lab (FPT University) in collaboration with SKKU, Kyung Hee University. This study provides the first systematic benchmark of CutMix strategies in Federated Semi-Supervised Learning, evaluating four variants (L, U, L+U, Mix) under different labeled ratios and heterogeneous client distributions. Our findings show that CutMix improves convergence and efficiency on homogeneous datasets, while FedFixMatch without CutMix remains highly competitive on heterogeneous data. The results offer practical guidelines for applying augmentation in real-world federated medical imaging scenarios with limited annotations.
,%20Macau,%20China.){kind=link}