Open Internet by MindsNet

Reasoning about Data Augmentations for Model Generalization

Choosing data augmentations for model generalization is a complex, heuristic process. The difficulty lies in reasoning about the assumptions each augmentation makes about the data, ensuring they are label-preserving, and determining their aggressiveness. This process is currently not systematic, often relying on intuition and leftover transforms from previous projects.

Computing & Technology, Computer Science, Machine Learning