Fine-tuning approach in TensorFlow - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is the fine-tuning approach in machine learning?

Fine-tuning is a method where you take a pre-trained model and continue training it on a new, often smaller dataset to adapt it to a specific task.

intermediate

Why do we freeze some layers during fine-tuning?

We freeze layers to keep the learned features intact and only update the later layers to adapt to the new task, which helps prevent overfitting and reduces training time.

beginner

In TensorFlow, how do you freeze layers of a pre-trained model?

You set the layer's 'trainable' attribute to False, for example: model.layers[i].trainable = False.

beginner

What is the benefit of using a pre-trained model for fine-tuning instead of training from scratch?

Pre-trained models have already learned useful features from large datasets, so fine-tuning saves time, requires less data, and often achieves better results.

intermediate

What is a common practice after freezing layers and before fine-tuning the model?

Compile the model again with a low learning rate to carefully adjust weights without large changes.

What does freezing layers in a model mean?

A Removing layers from the model

B Preventing layers from updating during training

C Adding new layers to the model

D Increasing the learning rate for some layers

Freezing layers means setting them so their weights do not change during training.

Why is fine-tuning useful when you have a small dataset?

A Because it trains the model from scratch

B Because it increases the dataset size automatically

C Because it ignores the small dataset

D Because it uses pre-learned features to adapt quickly

Fine-tuning leverages features learned from large datasets to work well on smaller datasets.

In TensorFlow, how do you make a layer non-trainable?

A layer.trainable = False

B layer.freeze()

C layer.trainable = True

D layer.stop_training = True

Setting layer.trainable = False freezes the layer during training.

What should you do after freezing layers before continuing training?

A Increase the batch size drastically

B Delete the optimizer

C Compile the model with a low learning rate

D Remove the frozen layers

Compiling with a low learning rate helps fine-tune the model gently.

Which of these is NOT a benefit of fine-tuning?

A Requires large new datasets

B Better performance on new tasks

C Faster training

D Uses pre-trained knowledge

Fine-tuning usually requires smaller datasets, not large ones.

Explain the steps involved in fine-tuning a pre-trained model using TensorFlow.

Think about what you keep, what you change, and how you train.

Describe why fine-tuning can improve model performance on a new task compared to training from scratch.

Consider the advantage of starting with knowledge already learned.