What is Transfer Learning

Surely someone has spend the time crafting the right model for your job...

• We can leverage an existing neural network architecture which is proven to work on problems similar to our own.
• Can leverage a working network architecture which has already learned patterns similar data to our own. (often results in great results with less data)

Transfer Learning is leveraging a working model's existing architecture and learned patterns for our own problem.

Things will be doing in this lecture :

• Introduce Transfer Learning with TensorFlow.
• Using a small dataset to experiment faster (10% training examples).
• Building a transfer learning feature extraction model with TensorFlow Hub.
• Use TensorBoard to track modelling experiments and results.

What are Callbacks?

• Callbacks are a tool which can add helpful functionality to your models during the training, evaluation or inference.

• Some Popular Callbacks include:

  • TensorBoard
  • ModelCheckpoint
  • EarlyStopping

  

Why do we want to track our experiments?

Earlier, to compare the results of one model's with other models we were scrolling up and down looking at the metrics, Oh boy that was tedious when you have 20 models or so.

Creating a TensorBoard Callback

tf.keras.callback.TensorBoard() —> to create a tensorboard callback