Feature extraction strategy in PyTorch - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is a feature extraction strategy in machine learning?

It is a method to transform raw data into useful information (features) that a model can learn from more easily.

intermediate

Why do we freeze layers in a pretrained model during feature extraction?

Freezing layers means we do not update their weights during training. This keeps the learned features intact and reduces training time.

beginner

In PyTorch, how do you freeze the parameters of a pretrained model?

You set each parameter's requires_grad attribute to False, like: <br>for param in model.parameters():<br>   param.requires_grad = False

intermediate

What is the difference between feature extraction and fine-tuning?

Feature extraction uses pretrained model layers as fixed feature detectors. Fine-tuning updates some or all pretrained layers to better fit new data.

intermediate

How can you replace the final layer of a pretrained model in PyTorch for feature extraction?

You assign a new layer to the model's classifier or fc attribute, for example:<br>model.fc = nn.Linear(in_features, num_classes)

What does freezing layers in a pretrained model do?

A Deletes the layers from the model

B Prevents their weights from updating during training

C Adds new layers to the model

D Increases the learning rate for those layers

Freezing layers means their weights stay fixed and do not change during training.

Which PyTorch attribute controls if a parameter is trainable?

A requires_grad

B trainable

C grad_enabled

D update_weights

The requires_grad attribute tells PyTorch whether to compute gradients for that parameter.

What is the main goal of feature extraction?

A To increase the size of the dataset

B To train a model from scratch

C To use learned features from a pretrained model to help a new task

D To remove irrelevant data points

Feature extraction leverages existing learned features to improve learning on new tasks.

When replacing the final layer in a pretrained model, what must you adjust?

A The optimizer type

B The input image size

C The learning rate only

D The number of output features to match your task classes

The final layer's output size must match the number of classes in your new task.

Which of these is NOT a benefit of feature extraction?

A Always achieves perfect accuracy

B Requires less data

C Faster training time

D Uses pretrained knowledge

Feature extraction helps but does not guarantee perfect accuracy.

Explain how to perform feature extraction using a pretrained model in PyTorch.

Think about which parts of the model you keep fixed and which parts you change.

Describe the difference between feature extraction and fine-tuning in transfer learning.

Consider how much of the pretrained model you allow to learn.