Post-processing improves generative AI outputs to match target attributes
Statistical attribute alignment for black-box generative AI via output post-processing
Artificial IntelligenceMachine Learning
Summary
It can be hard to make AI-generated content, like pictures or profiles, have specific characteristics that users want, such as fairness or diversity. The authors developed a way to adjust the outputs after they are created, without needing to change how the AI works internally. Their method uses many tries from the AI to pick outputs that better match a desired distribution of features. They tested their method on image and persona generation tasks and showed it complements other ways of guiding AI output.
What this means in practice
- •For marketing teams: Generate customer personas with attributes that match specific demographic targets for improved campaign relevance.
- •For advertising creatives: Produce images with attribute distributions aligned to advertising standards for better audience representation.
Authors
Kevin Jiang, Morgane Austern, Edgar Dobriban, Jason M. Klusowski
Abstract
Generative AI systems are increasingly used, but aligning their outputs with user requirements poses a continuing challenge. Here, we aim to ensure that the distribution of an attribute of an AI-generated output aligns with a user-specified target. This is motivated by examples such as fairness, where we want to ensure that a protected attribute (e.g., gender, race, or age categories) follows a desired distribution, and synthetic data generation, where we want the generated data to be representative of a target distribution. We study the practically important black-box access setting, where a user can repeatedly query a generative AI model. The goal is to return $m\ge 1$ outputs whose joint attribute distribution is as close as possible to this target. For both exact and approximate alignment, we develop algorithms that minimize the expected number of queries to the generator, and we further demonstrate their optimality as the number of requested outputs $m \rightarrow \infty$. Experiments on text-to-image generation and geocoded persona generation tasks show that our post-processing algorithms improve statistical attribute alignment, complementing prompting-based interventions.