Towards Modality-imbalanced Federated Graph Learning: A Data Synthesis-based Approach
Researchers have introduced FedMGS, a novel approach to address modality imbalance in federated graph learning. This method tackles challenges arising from missing modalities at both the client and node levels by synthesizing missing semantic representations within the latent space. FedMGS incorporates an availability-aware graph encoder, a prototype-guided latent semantic synthesizer, and a reliability-calibrated semantic fusion mechanism to improve performance and efficiency. Experiments demonstrate significant gains over existing baselines. AI
IMPACT This research could improve the robustness and applicability of federated learning models in real-world scenarios with incomplete data.