FedGUI: First Comprehensive Benchmark for Federated GUI Agents Across Mobile, Web, and Desktop

arXiv·medium signal

FedGUI introduces the first benchmark for developing and evaluating federated GUI agents that work across mobile, web, and desktop platforms. The benchmark provides six curated datasets to systematically evaluate cross-platform heterogeneity — a critical gap since centralized training of GUI agents faces significant cost and scalability challenges, while existing federated learning benchmarks don't capture real-world GUI diversity.