Implementing neural networks

You can find the code examples discussed in class in two google colab notebooks:

  1. A synthetic example, where we generate a data set and train a small fully connected network with ~17,000 parameters. The code runs reasonably well on a CPU or GPU.

  2. A real example, where we import a large pretrained model with ~ 11,000,000 parameters and retrain it on a new data set ('transfer learning'). You do want to use a GPU for this one.

Further details can be found in the notebooks.