The field of graph neural networks is moving towards addressing key challenges such as oversmoothing and oversquashing, with a focus on developing innovative architectures and techniques to improve performance. Researchers are exploring the use of Riemannian geometry, active diffusion, and graph rewiring to enhance the expressiveness and effectiveness of graph neural networks. Notably, the incorporation of physical symmetries and the development of novel network architectures are also being investigated to improve the generalization behavior of these models. Noteworthy papers include: Identifying internal patterns in (1+1)-dimensional directed percolation using neural networks, which demonstrates the capability of deep architectures to extract hierarchical structures from raw data. Deeper with Riemannian Geometry: Overcoming Oversmoothing and Oversquashing for Graph Foundation Models, which proposes a local approach to adaptively adjust message passing based on local structures, leading to improved expressiveness. An Active Diffusion Neural Network for Graphs, which achieves active diffusion by integrating multiple external information sources, effectively overcoming the over-smoothing problem.