Improving Fairness and Mitigating MADness in Generative Models
This addresses fairness and stability issues in generative models, which is important for applications requiring unbiased data generation, though it appears incremental as it builds on existing hypernetwork and regularization techniques.
The paper tackles unfairness, model autophagy disorder (MADness), and statistical bias in generative models by training them with intentionally designed hypernetworks and a regularization term, resulting in improved fairness for minority classes, stability in self-consumed settings, and less biased parameter estimates.
Generative models unfairly penalize data belonging to minority classes, suffer from model autophagy disorder (MADness), and learn biased estimates of the underlying distribution parameters. Our theoretical and empirical results show that training generative models with intentionally designed hypernetworks leads to models that 1) are more fair when generating datapoints belonging to minority classes 2) are more stable in a self-consumed (i.e., MAD) setting, and 3) learn parameters that are less statistically biased. To further mitigate unfairness, MADness, and bias, we introduce a regularization term that penalizes discrepancies between a generative model's estimated weights when trained on real data versus its own synthetic data. To facilitate training existing deep generative models within our framework, we offer a scalable implementation of hypernetworks that automatically generates a hypernetwork architecture for any given generative model.