Graph Convolutional Attention Reframes Graph Denoising Through a Spectral Lens

arXiv·low signal

This paper reframes attention-based graph denoising — the core operation of graph diffusion models — through a spectral perspective, connecting attention to graph convolution. The contribution is largely theoretical/architectural for graph-learning researchers, with limited near-term payoff for mainstream builders unless working directly on graph diffusion. Included as a low-importance signal for GNN practitioners.