EfficientEquivariantNetwork SupplementaryMaterials AMNIST-rotModelArchitecture

The hyperparameters we use in this architecture are kernel size k = 5, reduction ratior = 1, and the number of slicess = 2. In the large model, we increase the channel dimension to24, the number of slices to12, the reduction ratio to2, and keep other hyperparametersthesame. We take ResNet-18 [2], which is composed of an initial convolution layer, followed by 4 stage Res-Blocks and one final classification layer. The channel dimensions of each stage of Res-Blocks are 64 64 128 256 512. Thisindicates attentive G-Conv can be seen as the special cases of the Eqn.(25).