This Jupyter notebook shows a demo of classification on fashion mnist dataset(a data set containing items of clothing) using deep neural networks.
By using one hidden layer and output layer, we can get accuracy 89%, loss 0.2935 on training images and accuracy 86%, loss 0.3825 on test images.
After adding two (convolution + maxpool) layers, we can get accuracy 96%, loss 0.1032 on training images and accuracy 91%, loss 0.2277 on test images.