Pre-trained models (VGG, ResNet, MobileNet) in TensorFlow - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is a pre-trained model in machine learning?

A pre-trained model is a model that has already been trained on a large dataset. It can be reused to solve similar problems without training from scratch, saving time and resources.

beginner

What is the main advantage of using VGG, ResNet, or MobileNet pre-trained models?

They provide a strong starting point with learned features from large datasets like ImageNet, enabling faster training and often better accuracy on new tasks.

intermediate

How does ResNet improve deep neural networks compared to older models like VGG?

ResNet uses skip connections (or residual connections) that help avoid the problem of vanishing gradients, allowing very deep networks to train effectively.

intermediate

Why is MobileNet suitable for mobile and embedded devices?

MobileNet is designed to be lightweight and efficient, using depthwise separable convolutions to reduce computation and model size, making it ideal for devices with limited resources.

beginner

What is the typical use of pre-trained models in TensorFlow?

They are often used for transfer learning, where you take a pre-trained model, keep its learned features, and retrain the last layers on your own dataset for a new task.

Which dataset are VGG, ResNet, and MobileNet commonly pre-trained on?

A CIFAR-10

B ImageNet

C MNIST

D COCO

These models are commonly pre-trained on ImageNet, a large dataset with millions of labeled images across 1000 classes.

What key feature allows ResNet to train very deep networks effectively?

A Dropout layers

B Batch normalization

C Skip connections

D Pooling layers

Skip connections help gradients flow through the network, preventing vanishing gradient problems in deep networks.

MobileNet reduces computation mainly by using which technique?

A Depthwise separable convolutions

B LSTM layers

C Fully connected layers

D Data augmentation

Depthwise separable convolutions split convolution into two simpler steps, reducing computation and model size.

When using a pre-trained model for transfer learning, which part is usually retrained?

A Only the input layer

B The first convolutional layer

C All layers from scratch

D The last few layers

Typically, the last few layers are retrained to adapt the model to the new task while keeping earlier learned features.

Which pre-trained model is known for its very deep architecture with 16 or 19 layers?

A VGG

B ResNet

C MobileNet

D AlexNet

VGG models are famous for their simple and very deep architecture with 16 or 19 layers.

Explain how pre-trained models like VGG, ResNet, and MobileNet help in solving new image recognition tasks.

Think about how learning from a big example helps with smaller new problems.

Describe the main architectural difference between VGG and ResNet and why it matters.

Focus on how ResNet solves a problem that VGG faces with very deep layers.