Federated Prompt Learning: A Survey of Privacy-Preserving LLM Training

A new survey paper on arXiv (2608.13844) reviews recent advances in federated prompt learning (FPL), a technique that integrates federated learning (FL) with large language models (LLMs) to enable privacy-preserving training and reasoning. The paper addresses three research questions: the motivations and enabling technologies of FPL, its differences from conventional FL and full-model federated fine-tuning, and the trade-offs in performance, communication efficiency, and computational cost. The authors note that LLMs are core to cloud-based intelligent services but face challenges such as high computational costs, data centralization, and privacy concerns. FL offers a decentralized training paradigm that allows clients to collaboratively train models without sharing raw data, making it a promising solution. The survey aims to provide a comprehensive overview of FPL, highlighting its potential to reduce communication and computation overhead while maintaining model performance. The paper is categorized as a cross-type announcement on arXiv and is available at the provided URL.