Tag
This paper introduces Sheaf-based Federated Representation Learning (SFRL), a framework that aligns heterogeneous local representations via learnable sheaf restriction maps and a quadratic gluing regularizer, without assuming a shared global latent space. A decentralized algorithm (Sheaf-FRL) with convergence guarantees is proposed and shown to outperform baselines in cooperative classification under data and model heterogeneity.
MirrorWorld is a reflection-aware video inpainting framework that improves mirror reflection generation in videos by separately modeling semantic content (SRD) and geometric spatial arrangements (GTA), achieving better reflection reconstruction than existing image-based and video inpainting baselines.