Prompt Public Large Language Models to Synthesize Data for Private On-device Applications

arXiv:2404.04360v1 Announce Type: cross
Abstract: Pre-training on public data is an effective method to improve the performance for federated learning (FL) with differential privacy (DP). This paper investigates how large language models (LLMs) trained on public data can improve the quality of pre-training data for the on-device language models trained with DP and FL. We carefully design LLM prompts to filter and transform existing public data, and generate new data to resemble the real user data distribution. The model pre-trained …

applications arxiv can cs.cl cs.cr cs.lg data device differential privacy federated federated learning language language models large llms performance privacy private prompt public quality training training data