cnn-extensions.md

Next Steps with CNNs

1. Build a SavedModel so that you can easily load your model and use it for inference later.
2. In a new notebook, load your SavedModel and try using it for inference.
3. Refactor you model to use dropout during training. Look into tf.layers.dropout.
4. Try creating a deeper network. How much does each layer decrease training speed? Think about ways to benchmark training times.
5. How much faster is your p3 instance at training than your laptop? Compare training speeds.