F = 8 for SVHN and MNIST and F = 32 for the other experiments. Optimization: We optimized using SGD with ADAM [17]. For all datasets we used a learning rate ...
確定! 回上一頁