Efficient Fairness Auditing Across Guidance Scales in Text-to-Image Diffusion Models via Causal Abstraction
A causal-abstraction-based audit instrument for efficiently evaluating fairness under interventions on the classifier-free guidance scale is proposed and experiments evaluate distributional fidelity, fairness-query accuracy, and computational efficiency.
Nabila Tasfiha Rahman, Rajatsubhra Chakraborty, De-Peng Xu et al.
· 0 citations