Noise-Resilient Federated Learning: Suppressing Noisy Labels in the Local Datasets of Participants

Rahul Mishra, Hari Prabhat Gupta, Tanima Dutta

Abstract: Federated Learning (FL) is a novel paradigm of collaboratively training a model using local datasets of multiple participants. FL maintains data privacy and keeps local datasets confined to the participants. This poster presents a novel noiseresilient federated learning approach that suppresses the negative impact of noisy labels in the local datasets of the participants. The approach starts with the estimation of noise ratio without using prior information about the concentration of noisy labels. Next, the server generates different groups of participants using the estimated noise ratio. The FL-based training starts with the group having the least noise ratio, and subsequent groups are added later. We also introduce a noise robust loss function that incorporates dynamic variables to reduce the impact of noisy labels. The proposed approach reduces the overall training time and achieves adequate accuracy despite noisy labels.