On May 25th, the reporter learned from Shanxi University that the team of the Institute of Intelligent Information Processing at the university used graph controlled networks to achieve high-performance point cloud segmentation. The related results were published in the international journal IEEE Pattern Analysis and Machine Intelligence in the field of artificial intelligence. Point cloud data analysis is widely used in fields such as autonomous driving, 3D understanding, and robotics. Point cloud segmentation is a fundamental and challenging task in the field of point cloud data analysis, aimed at dividing the target point cloud into different regions based on different attributes and functions. "The key to achieving high-performance point cloud segmentation is to extract discriminative point by point features." A team member from the Institute of Intelligent Information Processing at Shanxi University explained that there is often strong heterogeneity among neighboring nodes at the boundaries of different segmentation regions in point clouds. The previous method ignored the homogeneous and heterogeneous relationships between nodes in the process of implementing feature aggregation. This causes unnecessary heterogeneous node information to be mixed in with node features, resulting in blurred boundary of point cloud segmentation. In response to the above issues, the research team proposed a graph regulatory network that models point clouds as homogeneous heterogeneous graphs. They combined the graph attention model and designed a graph attention convolution based on homogeneity guidance, mining homogeneous features within local neighborhoods. The research team further designed a prototype feature extraction module to further explore homogeneous features from the global prototype space, improve node feature discrimination, and further enhance point cloud segmentation performance. It is reported that the research results have further enhanced the discriminability of node features and improved the clarity of point cloud segmentation boundaries. (Lai Xin She)