Search papers, labs, and topics across Lattice.
The paper introduces Fed-GAME, a federated learning framework for time-series forecasting that addresses client heterogeneity by learning a dynamic implicit graph for personalized aggregation. Clients transmit parameter differences between their local and the global model, which are then processed by a Graph Attention Mixture-of-Experts (GAME) aggregator on the server. This aggregator uses shared experts to score and personalized gates to weight selective updates, enabling fine-grained personalization.
Personalized federated learning gets a boost from Fed-GAME, which uses a graph attention mixture-of-experts to dynamically aggregate client updates, outperforming SOTA methods on time-series forecasting tasks.
Federated learning (FL) on graphs shows promise for distributed time-series forecasting. Yet, existing methods rely on static topologies and struggle with client heterogeneity. We propose Fed-GAME, a framework that models personalized aggregation as message passing over a learnable dynamic implicit graph. The core is a decoupled parameter difference-based update protocol, where clients transmit parameter differences between their fine-tuned private model and a shared global model. On the server, these differences are decomposed into two streams: (1) averaged difference used to updating the global model for consensus (2) the selective difference fed into a novel Graph Attention Mixture-of-Experts (GAME) aggregator for fine-grained personalization. In this aggregator, shared experts provide scoring signals while personalized gates adaptively weight selective updates to support personalized aggregation. Experiments on two real-world electric vehicle charging datasets demonstrate that Fed-GAME outperforms state-of-the-art personalized FL baselines.
