Understanding data augmentation for classification: when to warp?
This work addresses the problem of selecting effective data augmentation strategies for machine learning practitioners, but it is incremental as it builds on existing augmentation methods.
The paper investigates the effectiveness of data augmentation for classification, comparing data warping (transformations in data-space) and synthetic over-sampling (in feature-space) on MNIST dataset with neural networks, SVMs, and extreme learning machines. It finds that data-space augmentation yields greater performance improvements and reduces overfitting when plausible transforms are known.
In this paper we investigate the benefit of augmenting data with synthetically created samples when training a machine learning classifier. Two approaches for creating additional training samples are data warping, which generates additional samples through transformations applied in the data-space, and synthetic over-sampling, which creates additional samples in feature-space. We experimentally evaluate the benefits of data augmentation for a convolutional backpropagation-trained neural network, a convolutional support vector machine and a convolutional extreme learning machine classifier, using the standard MNIST handwritten digit dataset. We found that while it is possible to perform generic augmentation in feature-space, if plausible transforms for the data are known then augmentation in data-space provides a greater benefit for improving performance and reducing overfitting.