Google Cloud AutoML is pretty cool! I read the related paper this afternoon: Learning Transferable Architectures for Scalable Image Recognition by Barret et al. https://lnkd.in/gb2ytXw It is a smart way to use NAS training two kinds of convolutional cells(layers) and stack them together to do image classification on large data set. It avoids training NAS on large dataset directly which is time/computation consuming. Learning network will be a really exciting direction to explore.