Can Public Large Language Models Help Private Cross-device Federated Learning?
This work addresses the challenge of training private, on-device language models efficiently, which is incremental but important for applications requiring user-level differential privacy in federated settings.
The paper tackles the problem of improving privacy-utility trade-offs in differentially private federated learning of language models by leveraging public large language models and data, achieving significant improvements through techniques like distillation and a novel distribution matching algorithm.
We study (differentially) private federated learning (FL) of language models. The language models in cross-device FL are relatively small, which can be trained with meaningful formal user-level differential privacy (DP) guarantees when massive parallelism in training is enabled by the participation of a moderate size of users. Recently, public data has been used to improve privacy-utility trade-offs for both large and small language models. In this work, we provide a systematic study of using large-scale public data and LLMs to help differentially private training of on-device FL models, and further improve the privacy-utility tradeoff by techniques of distillation. Moreover, we propose a novel distribution matching algorithm with theoretical grounding to sample public data close to private data distribution, which significantly improves the sample efficiency of (pre-)training on public data. The proposed method is efficient and effective for training private models by taking advantage of public data, especially for customized on-device architectures that do not have ready-to-use pre-trained models.