Optimizing and Accelerating Graph Neural Networks for Large- Scale Irregular IoT Sensor Data on Chinese NPU Devices

This project utilizes distributed training of graph neural networks (GNNs) on Chinese NPUs to process vast, irregularly sampled IoT time-series data, enhancing analysis efficiency. Key novelties include integrating attention mechanisms with locality-sensitive hashing on missing data imputation, optimizing distributed processing on domestic edge devices, and achieving high training efficiency and low deployment costs with Chinese NPUs.