using the adjacency matrix to represent the connection of different operations. Only convolution max-pooling 33# are allowed to be used to build the neural architectures. The best architecture achieves a mean test error of 5.68%, and the mean test error of the architecture with the best validation error is 5.77%. b) NASBench-201 NASBench-201 [19] is a recently proposed NAS benchmark, and it contains 15.6k trained architectures for image classification. Each architecture is trained once on the CIFAR-10, CIFAR-100 [46], and ImageNet-16-120, and ImageNet-16-120 is a downsampled variant of ImageNet [47]. The structure of each architecture and its evaluation details such as training error, validation error, and test error of each architecture are reported. Each architecture