Chapter 6 – Data Imbalance in Deep Learning

1. The main challenge stems from the different types of data these models handle. Classical machine learning models typically work with structured, tabular data, while deep learning models handle unstructured data such as images, text, audio, and video.
2. An imbalanced version of the MNIST dataset can be created by randomly selecting a certain percentage of examples for each class. This process involves choosing indices of the samples to remove and then actually removing these samples from the training set.
3. This has been left as an exercise for you.
4. Random oversampling is used to address imbalance in the dataset. It works by duplicating samples from the minority classes until each class has an equal number of samples. This technique is usually considered to perform better than no sampling.
5. Data augmentation techniques can include rotating, scaling, cropping, blurring, adding noise to the image, and much more. However...