model collapse

Model collapse refers to a phenomenon where machine learning models, particularly generative models, degrade over time due to training on synthetic data, including outputs from previous versions of themselves.