FairGen : Controlling Sensitive Attributes for Fair Generations in Diffusion Models via Adaptive Latent Guidance

Mintong Kang, Vinayshekhar Bannihatti Kumar, Shamik Roy, Abhishek Kumar, Sopan Khosla, Balakrishnan Murali Narayanaswamy, Rashmi Gangadharaiah · 2025

Read FairGen : Controlling Sensitive Attributes for Fair Generations in Diffusion Models via Adaptive Latent Guidance by Mintong Kang, Vinayshekhar Bannihatti Kumar, Shamik Roy, Abhishek Kumar, Sopan Khosla, Balakrishnan Murali Narayanaswamy, Rashmi Gangadharaiah (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-01872.txt.

Description

Mintong Kang and six co-authors present a fairness tool for image generators in a 2025 paper. Text to image models often skew who they depict. A prompt for an engineer, say, may yield mostly men. Such skew raises ethical worries and slows adoption.

Their method, FairGen, steers the generation process as each image forms. A guidance module nudges hidden states toward the wanted attribute mix. A memory module tracks output counts and corrects the steering. The team also built a broader bias test with richer prompts. On standard tests, gender bias fell by more than two thirds. Image quality held up under the steering in their checks. The tool can target any attribute value a user sets.

Catalog record

CREATORMintong Kang, Vinayshekhar Bannihatti Kumar, Shamik Roy, Abhishek Kumar, Sopan Khosla, Balakrishnan Murali Narayanaswamy, Rashmi Gangadharaiah
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-01872

Browse the full library · About the Tower of Babel