Universally Invariant Learning in Equivariant GNNs

Deli Zhao (Xiaomi AI Lab) · Yu Rong (Shenzhen Tencent Computer System Co., Ltd.) · Tingyang Xu (Tencent AI Lab) · Anyi Li (Renmin University of China) · Jiacheng Cen (Renmin University of China) · Wenbing Huang (Tsinghua University) · Ning Lin (Renmin University of China) · Zihe Wang (Renmin University of China)
canonical formcomplete scalar functioncomputational costdeeper architecturesefficient algorithmegnnempirical resultsequivariant graph neural networksfull-rank steerable basis setgeometric graphmulti-body interactionspolynomial-time solutionssteerable featurestfnuniversal approximation property

Equivariant Graph Neural Networks (GNNs) have demonstrated significant success across various applications. To achieve completeness