1. Official Vendor Documentation (AWS): Vamshi, K., & Nick, L. (2021, June 29). Handle class imbalance with Amazon SageMaker. AWS Machine Learning Blog. In the "Data-level methods" section, the authors state, "Data-level methods, also known as resampling, modify the training dataset to balance the class distribution... Oversampling methods add minority class samples to balance the dataset... More complex methods generate synthetic samples." This directly supports using data augmentation (generating synthetic samples) to fix imbalanced classes.
2. University Courseware: Stanford University. (n.d.). CS231n: Convolutional Neural Networks for Visual Recognition - Lecture 5. Retrieved from https://cs231n.github.io/convolutional-networks/. The section on "Data Augmentation" explains that this technique artificially enlarges the dataset by creating modified copies of existing data. This is the fundamental principle used to increase the representation of minority classes to combat bias.
3. Official Vendor Documentation (AWS): Amazon SageMaker Developer Guide. Data Augmentation. The guide details how data augmentation "helps prevent the model from overfitting and helps the model generalize better." While often cited for generalization, this process of creating varied data is the same mechanism used to balance datasets and reduce sampling bias.