Unit 5 - Training Convolutional Neural Networks

What is the purpose of data normalization when preparing a dataset for training a CNN?

• To make the data larger and more complex
• To make the data smaller and easier to manage
• To scale and center the data to a standard range
• To add noise to the dataset

In CNN data preparation, what is the role of data augmentation techniques?

• To create entirely new synthetic data
• To reduce the size of the dataset
• To increase the dataset's variability by applying transformations to the existing data
• To remove noisy data points

When splitting a dataset into training, validation, and test sets for CNN training, what is the typical ratio?

• 50% training, 25% validation, 25% test
• 70% training, 20% validation, 10% test
• 80% training, 10% validation, 10% test
• 60% training, 30% validation, 10% test

Which type of labeling strategy is used when each data point in a dataset belongs to a single, mutually exclusive category?

• Multi-labeling
• Binary labeling
• Multiclass labeling
• Hierarchical labeling

When dealing with an imbalanced dataset for CNN training, which of the following techniques can help address this issue?

• Removing the majority class data
• Oversampling the minority class data
• Randomly shuffling the dataset
• Ignoring the class imbalance and training as usual

Which file format is commonly used for storing and organizing large image datasets?

• .txt
• .csv
• .jpg
• .tfrecord

What is the purpose of stratified sampling when splitting a dataset into training and testing subsets?

• To ensure each class in the dataset is proportionally represented in both subsets
• To increase the dataset size
• To randomly shuffle the data
• To remove outliers from the dataset

What is one of the main advantages of using data pipelines in CNN training?

• Data pipelines are not necessary for CNN training.
• They allow for better compression of the dataset.
• They enable efficient data loading and preprocessing on-the-fly, reducing memory usage.
• They make it easier to overfit the model.

In transfer learning with CNNs, what does "fine-tuning" typically involve?

• Training the entire model from scratch
• Applying pre-trained weights directly to the new task
• Modifying only the output layer and training the rest of the model
• Ignoring the pre-trained weights

Which technique is used to handle missing data when preparing a dataset for CNN training?

• Replacing missing data with random values
• Deleting data points with missing values
• Ignoring missing data and proceeding with training
• Imputing missing values using statistical methods