POSTER: Privacy-preserving Federated Active Learning

Active learning is a technique for investigating a way to maximize performance with minimal labeling effort and let the machine automatically and adaptively selects the most informative data for labeling. Since the labels on records may contain sensitive information, a privacy-preserving active learning scheme was proposed by applying differential privacy. Another type of privacy-preserving machine learning as federated learning should be considered, which is a distributed machine learning framework providing the protection of client data. We propose an encryption-based Federated Learning approach to protect privacy in Active Learning. The experimental result shows a homomorphic encryption-based federated learning scheme can preserve privacy in active learning while keeping accuracy.