Transfer Learning and Pre-trained CNNs Flashcards
What is a pretrained model?
A pretrained model is a saved neural network that has been previously trained on a large dataset and can be reused for new tasks.
Why are pretrained models useful?
They improve accuracy and save training time, especially when working with small datasets.
What is ImageNet?
ImageNet is a large-scale dataset with 1.4 million labeled images and 1,000 different classes.
Name three examples of pretrained models.
VGG-16, GoogLeNet (Inception), and MobileNet.
What is VGG-16?
A deep convolutional neural network developed by the VGG group at the University of Oxford, trained on ImageNet.
What is GoogLeNet (Inception)?
A deep CNN introduced by Google in 2014, known for using inception modules to improve efficiency.
What is MobileNet?
A CNN designed for mobile and edge devices, optimized for efficiency.
What is Faster R-CNN used for?
Faster R-CNN is used for object detection.
What is Mask R-CNN?
An extension of Faster R-CNN that provides pixel-wise segmentation masks along with bounding boxes.
What does YOLO stand for, and what is it used for?
YOLO (You Only Look Once) is an object detection algorithm that divides an image into a grid and predicts bounding boxes and class probabilities.
What are two main ways to use a pretrained model?
Feature extraction and fine-tuning.
What is feature extraction in transfer learning?
Using a pretrained model’s convolutional base to extract features from new data and feeding them into a new classifier.
What is fine-tuning in transfer learning?
Unfreezing some layers of a pretrained model and training them on new data to adapt the model to a new task.
Why might you reuse only the convolutional base of a pretrained model?
Because it captures general image features that can be useful for different tasks, while the classifier is often task-specific.
What is shallow mode in transfer learning?
Running the convolutional base over a dataset, recording the outputs, and training a separate classifier on the extracted features.
What is deep mode in transfer learning?
Extending a pretrained model by adding new Dense layers on top and training the entire network end-to-end.
What is the key difference between shallow and deep mode?
Shallow mode uses the pretrained model as a fixed feature extractor, while deep mode trains additional layers on top.
What function in Keras loads the VGG16 convolutional base?
VGG16(weights='imagenet', include_top=False, input_shape=(224, 224, 3))
How do you freeze the convolutional base in Keras?
conv_base.trainable = False
How do you fine-tune specific layers in Keras?
Iterate through layers and set layer.trainable = True for the ones you want to train.
Name three additional pretrained models available in Keras.
Xception, ResNet50, and MobileNet.
What is the self-study reading list for transfer learning?
Chapter 6 of ‘Convolutional Neural Networks’ and papers by Simonyan & Zisserman (2014) and Szegedy et al. (2015).
