Architecture of AstroNet-Triage (identical to that of the best-performing neural network from Shallue & Vanderburg 2018). Convolutional layers are denoted conv 〈kernel size〉-〈number of feature maps〉, max pooling layers are denoted maxpool 〈window length〉-〈stride length〉, and fully connected layers are denoted Fully connected-〈number of units〉.