PRISM: Reducing Spurious Implicit Biases in Vision-Language Models with LLM-Guided Embedding Projection

TL;DR

PRISM is a novel, data-free method that reduces biases in vision-language models by using LLM-generated scene descriptions and a contrastive loss to learn bias-minimizing embeddings, improving fairness without external data.

Contribution

It introduces PRISM, a task-agnostic, data-free debiasing approach leveraging LLMs and a contrastive loss to mitigate spurious biases in vision-language models.

Findings

PRISM outperforms existing debiasing methods on Waterbirds and CelebA datasets.

It effectively minimizes spurious correlations while maintaining image-text alignment.

The method is applicable without relying on bias annotations or external data.

Abstract

We introduce Projection-based Reduction of Implicit Spurious bias in vision-language Models (PRISM), a new data-free and task-agnostic solution for bias mitigation in VLMs like CLIP. VLMs often inherit and amplify biases in their training data, leading to skewed predictions. PRISM is designed to debias VLMs without relying on predefined bias categories or additional external data. It operates in two stages: first, an LLM is prompted with simple class prompts to generate scene descriptions that contain spurious correlations. Next, PRISM uses our novel contrastive-style debiasing loss to learn a projection that maps the embeddings onto a latent space that minimizes spurious correlations while preserving the alignment between image and text embeddings.Extensive experiments demonstrate that PRISM outperforms current debiasing methods on the commonly used Waterbirds and CelebA datasets We make our code public at: https://github.com/MahdiyarMM/PRISM.