让边缘设备高效个性化训练生成模型，通信量减少99%。
Parameter-Efficient and Personalized Federated Training of Generative Models at the Edge
- 只联邦更新轻量适配器，冻结主模型降低计算压力
- 在PTB和CIFAR-10上实现更低困惑度与FID值，收敛更快
- 适合资源受限的异构边缘设备，支持本地个性化
大型生成模型（如语言和扩散模型）虽能生成高质量文本与图像，但在跨设备联邦设置中难以训练或适配，因计算与通信开销大，且存在统计与系统异构性。我们提出FedGen-Edge框架，将冻结的预训练全局主干与轻量级客户端适配器解耦，仅联邦传播适配器。采用低秩适应（LoRA）将客户端更新限制在紧凑子空间，相比全模型FedAvg，上行通信量减少超99%，在非独立同分布数据下稳定聚合，并天然支持个性化——每个客户端可保留本地调优的适配器。在语言建模（PTB）与图像生成（CIFAR-10）任务中，FedGen-Edge比强基线取得更低困惑度与FID值，收敛更快，同时保持简单的FedAvg式服务器。简要消融实验显示，超过中等LoRA秩后收益递减，且本地训练轮次与客户端漂移间存在权衡。FedGen-Edge为异构边缘设备上的隐私保护、资源感知与个性化生成式AI提供了可行路径。
原文摘要 · Abstract (English)
Large generative models (for example, language and diffusion models) enable high-quality text and image synthesis but are hard to train or adapt in cross-device federated settings due to heavy computation and communication and statistical/system heterogeneity. We propose FedGen-Edge, a framework that decouples a frozen, pre-trained global backbone from lightweight client-side adapters and federates only the adapters. Using Low-Rank Adaptation (LoRA) constrains client updates to a compact subspace, which reduces uplink traffic by more than 99 percent versus full-model FedAvg, stabilizes aggregation under non-IID data, and naturally supports personalization because each client can keep a locally tuned adapter. On language modeling (PTB) and image generation (CIFAR-10), FedGen-Edge achieves lower perplexity/FID and faster convergence than strong baselines while retaining a simple FedAvg-style server. A brief ablation shows diminishing returns beyond moderate LoRA rank and a trade-off between local epochs and client drift. FedGen-Edge offers a practical path toward privacy-preserving, resource-aware, and personalized generative AI on heterogeneous edge devices.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。