Bias Begets Bias: The Impact of Biased Embeddings on Diffusion Models

TL;DR

This paper investigates how biased embedding spaces influence the fairness and evaluation of diffusion models in Text-to-Image systems, proposing fairness criteria, analyzing bias impacts, and suggesting mitigation strategies.

Contribution

It introduces statistical fairness criteria based on internal representations, demonstrates the necessity of unbiased embeddings for balanced diffusion models, and develops a framework for bias mitigation in evaluation.

Findings

Biased embeddings lead to unfair image generation and lower alignment scores.

Unbiased embedding spaces are necessary for fair and diverse diffusion outputs.

Proposed bias mitigation methods improve fairness in diffusion model evaluations.

Abstract

With the growing adoption of Text-to-Image (TTI) systems, the social biases of these models have come under increased scrutiny. Herein we conduct a systematic investigation of one such source of bias for diffusion models: embedding spaces. First, because traditional classifier-based fairness definitions require true labels not present in generative modeling, we propose statistical group fairness criteria based on a model's internal representation of the world. Using these definitions, we demonstrate theoretically and empirically that an unbiased text embedding space for input prompts is a necessary condition for representationally balanced diffusion models, meaning the distribution of generated images satisfy diversity requirements with respect to protected attributes. Next, we investigate the impact of biased embeddings on evaluating the alignment between generated images and prompts, a process which is commonly used to assess diffusion models. We find that biased multimodal embeddings like CLIP can result in lower alignment scores for representationally balanced TTI models, thus rewarding unfair behavior. Finally, we develop a theoretical framework through which biases in alignment evaluation can be studied and propose bias mitigation methods. By specifically adapting the perspective of embedding spaces, we establish new fairness conditions for diffusion model development and evaluation.