Deep learning technology,like convolutional neural net-works,has been widely used in machine learning appli-cations such as image detection and classification,speech processing,natural language processing and cross-modal re-trieval[1].Although deep learning technology has achieved great success in various fields,it has been unable to match non-Euclidean data effectively.In this context,the emer-gence of geometric deep learning has filled the above-mentioned technical gaps and realized an effective combi-nation of deep learning technology and non-Euclidean data,consisting of manifolds and graphs.Non-Euclidean data is very common,so it is important to study non-Euclidean data with deep learning.