Episode 8 (optional): using a GPU

[svenvanderburg]

Rationale

Quickly show that using a GPU for training a network is both easy and speeds up training significantly.

Learning objectives

• Use a GPU when training a neural network.
• Describe why using a GPU speeds up training significantly.

Questions

• Why would I use a GPU when training a network?
• How do set Keras to use the GPU on my machine?

Keypoints

• When applying state-of-the-art deep learning techniques you need to use a GPU for training.

Outline

• Train a state-of-the-art network on a dataset on CPU, and let learners experience that this is too slow.
• Show how to enable GPU in keras
• Train the same network on the same dataset with GPU.
• Explain (superficially) why using a GPU speeds up learning so much
• Provide context and links to resources for setting this up on different systems (i.e. AWS, google cloud, DAS, SURF cloud)

[svenvanderburg]

We could use google colab, it provides free access to a notebook on a GPU machine.