Framelet Transform

Framelet transforms are multi-scale signal processing techniques increasingly used to enhance graph neural networks (GNNs) and image processing. Current research focuses on applying framelets to address challenges like over-smoothing in deep GNNs, improving image deblurring and denoising, and enabling efficient signal processing on various data structures, including graphs and spherical signals. These advancements lead to improved performance in diverse applications such as medical imaging (e.g., BNCT dose reconstruction), graph classification, and signal processing tasks, offering more robust and efficient algorithms. The resulting models often leverage architectures like U-Net and incorporate techniques such as fractional calculus and energy-based regularization.