Latent Directions: A Simple Pathway to Bias Mitigation in Generative AI

TL;DR

This paper presents a simple, effective method for bias mitigation in generative AI by learning latent directions in the diffusion process, enabling diverse and inclusive image synthesis without altering prompts or embeddings.

Contribution

Introduces a novel latent space direction learning approach for bias mitigation in generative models, allowing linear combination and integration with text embedding adjustments.

Findings

Successfully mitigates geographical biases in generated images

Enables linear combination of latent directions for new bias mitigations

Provides a tool for developers to select concepts for bias mitigation

Abstract

Mitigating biases in generative AI and, particularly in text-to-image models, is of high importance given their growing implications in society. The biased datasets used for training pose challenges in ensuring the responsible development of these models, and mitigation through hard prompting or embedding alteration, are the most common present solutions. Our work introduces a novel approach to achieve diverse and inclusive synthetic images by learning a direction in the latent space and solely modifying the initial Gaussian noise provided for the diffusion process. Maintaining a neutral prompt and untouched embeddings, this approach successfully adapts to diverse debiasing scenarios, such as geographical biases. Moreover, our work proves it is possible to linearly combine these learned latent directions to introduce new mitigations, and if desired, integrate it with text embedding adjustments. Furthermore, text-to-image models lack transparency for assessing bias in outputs, unless visually inspected. Thus, we provide a tool to empower developers to select their desired concepts to mitigate. The project page with code is available online.