A convolutional neural network trained with mock galaxy cluster maps is applied to real maps from the Planck satellite, successfully predicting the masses of over 1,000 observed clusters and finding a 15% bias in masses measured directly from Planck.