Outputs of generative diffusion models are often unattributable.

Abstract

Modern generative diffusion models work by replicating the statistical patterns of large training datasets. Developing a method to attribute generated outputs to influential training data would greatly advance our understanding of and ability to regulate these models, leading to much work towards this goal. But is this possible? Here, we show that models trained with enough data often generate samples that are unattributable. We establish this through a large-scale analysis of what-if scenarios, revealing that we can often omit any sample or creator from the training data without affecting a generated sample. Our study focuses on diffusion models, which has become the dominant model for generating audiovisual media, and is also prevalent in many scientific applications including protein structure modeling and therapeutic discovery. Central to our analysis is a model ablation methodology that allows efficient removal of training examples from a trained model without the need to retrain.