Home

Welcome to the Transfer Learning 101 workshop.

There's several ways you can follow this workshop. It is based around a simple classification problem: real enough that images are big enough you don't need to squint (and aren't MNIST), but light enough that training takes only a few minutes even on a CPU (i7).

Either way, you can go one of two routes:

1. Looking & light tinkering

Start a Tensorboard, run the code in different ways following the guide in this this same wiki and see what changing certain parameters does.

2. Tinkering, breaking and rebuilding

Or you can write the code yourself, following the commits on the branch "workshop". For example, you can start at tag workshop_start:

git checkout workshop_start

If you get stuck, you can always checkout the next commit on branch "workshop" and carry on from there.

Notice there have been major changes in the last few months in the Tensorflow front. Keras is not the recommended API to use, so we'll use tensorflow.keras as much as possible and export the models to Tensorflow so they can be used with Tensorflow Serving, TFLite, etc.

A good place to start is the Tensorflow Guide. Notice the first recommendation, the tf.Keras API: that's where we'll start today.