Local K-Similarity Constraint for Federated Learning with Label Noise
Published in IEEE International Symposium on Biomedical Imaging (ISBI) 2026, 2025
This study introduces a local regularization objective for federated learning with noisy labels. The method uses the representation space of a self-supervised pretrained model to evaluate closeness among examples and enforce similarity between nearby samples within each client. This local constraint improves robustness when many heterogeneous clients contain noisy labels, helping prevent noisy clients from corrupting the global model.
Citation: Amgain, S., Shrestha, P., Khanal, B., Devkota, A., Shrestha, Y.R., Baek, S., Gyawali, P. and Bhattarai, B., 2026. Local K-Similarity Constraint for Federated Learning with Label Noise. IEEE International Symposium on Biomedical Imaging (ISBI).
Citation: Amgain, S., Shrestha, P., Khanal, B., Devkota, A., Shrestha, Y.R., Baek, S., Gyawali, P. and Bhattarai, B., 2026. Local K-Similarity Constraint for Federated Learning with Label Noise. IEEE International Symposium on Biomedical Imaging (ISBI).
Download Paper