Self-Correcting Self-Consuming Loops for Generative Model Training
This addresses a critical issue for AI researchers and practitioners working with generative models, as it enables stable training with synthetic data, though it is incremental by building on existing concepts of correction functions.
The paper tackles the problem of training instability and collapse in generative models when using synthetic data, known as self-consuming loops, by proposing a self-correction function that maps data points to be more likely under the true distribution, and demonstrates that this approach avoids model collapse even with a 100% synthetic-to-real data ratio in human motion synthesis.
As synthetic data becomes higher quality and proliferates on the internet, machine learning models are increasingly trained on a mix of human- and machine-generated data. Despite the successful stories of using synthetic data for representation learning, using synthetic data for generative model training creates "self-consuming loops" which may lead to training instability or even collapse, unless certain conditions are met. Our paper aims to stabilize self-consuming generative model training. Our theoretical results demonstrate that by introducing an idealized correction function, which maps a data point to be more likely under the true data distribution, self-consuming loops can be made exponentially more stable. We then propose self-correction functions, which rely on expert knowledge (e.g. the laws of physics programmed in a simulator), and aim to approximate the idealized corrector automatically and at scale. We empirically validate the effectiveness of self-correcting self-consuming loops on the challenging human motion synthesis task, and observe that it successfully avoids model collapse, even when the ratio of synthetic data to real data is as high as 100%.