The recent advancements in graph neural networks (GNNs) have shown significant progress in various aspects, including joint graph and sampling set selection, shape error prediction in machining, adversarial robustness evaluation, graph rationalization, and prediction uncertainty calibration. A notable trend is the integration of graph learning and signal sampling, exemplified by methods that jointly optimize graph structure and sampling set, enhancing reconstruction performance and reducing complexity. Another key development is the application of GNNs to predict shape errors in complex machining processes, leveraging graph structures to model spatial and temporal connections, which improves generalization and handles low label scenarios effectively. Adversarial robustness evaluation has emerged as a critical tool for improving GNN performance by identifying and focusing on robust nodes, thereby enhancing classification accuracy. Additionally, innovative approaches to graph rationalization have been proposed to boost environment diversity, ensuring more robust generalization across different distributions. Lastly, advancements in prediction uncertainty have led to more precise calibration techniques, grouping nodes by both neighborhood similarity and confidence levels for fine-grained calibration, resulting in significant error reductions. These developments collectively push the boundaries of GNN capabilities, making them more versatile and reliable for a wider range of applications.