Differentially Private Meta-Learning
This work addresses privacy concerns for task-owners in meta-learning applications like federated and few-shot learning, offering a practical relaxation that enhances performance while protecting data.
The paper tackles the privacy risks in parameter-transfer meta-learning by introducing task-global differential privacy as a relaxation of stricter models, and proposes a new algorithm that maintains privacy while showing improved performance in federated learning and few-shot classification tasks, with dramatic gains in neural language modeling and image classification.
Parameter-transfer is a well-known and versatile approach for meta-learning, with applications including few-shot learning, federated learning, and reinforcement learning. However, parameter-transfer algorithms often require sharing models that have been trained on the samples from specific tasks, thus leaving the task-owners susceptible to breaches of privacy. We conduct the first formal study of privacy in this setting and formalize the notion of task-global differential privacy as a practical relaxation of more commonly studied threat models. We then propose a new differentially private algorithm for gradient-based parameter transfer that not only satisfies this privacy requirement but also retains provable transfer learning guarantees in convex settings. Empirically, we apply our analysis to the problems of federated learning with personalization and few-shot classification, showing that allowing the relaxation to task-global privacy from the more commonly studied notion of local privacy leads to dramatically increased performance in recurrent neural language modeling and image classification.