DPAUC: Differentially Private AUC Computation in Federated Learning

Federated learning (FL) has gained significant attention recently as a privacy-enhancing tool to jointly train machine model by multiple participants. The prior work on FL mostly studied how protect label privacy during training. However, evaluation in might also lead the potential leakage of private information. In this work, we propose an algorithm that can accurately compute widely used AUC (area under curve) metric when using differential (DP) FL. Through extensive experiments, show our algorithms accurate AUCs compared ground truth. code is available at https://github.com/bytedance/fedlearner/tree/master/example/privacy/DPAUC