MNIST_lessParameter_10k Less parameters will improve the training time. FC layers are costly and we can avoid it if needed.