FairPIVARA: Reducing and Assessing Biases in CLIP-Based Multimodal Models

TL;DR

This paper introduces FairPIVARA, a method to reduce biases in CLIP-based multimodal models, demonstrating significant bias mitigation and improved fairness in zero-shot tasks, especially for Portuguese language applications.

Contribution

The paper presents FairPIVARA, a novel bias reduction technique for CLIP models that effectively diminishes discriminatory biases by removing affected feature dimensions.

Findings

Bias reduction of up to 98% achieved

More balanced word distribution in models

Effective bias mitigation in zero-shot tasks

Abstract

Despite significant advancements and pervasive use of vision-language models, a paucity of studies has addressed their ethical implications. These models typically require extensive training data, often from hastily reviewed text and image datasets, leading to highly imbalanced datasets and ethical concerns. Additionally, models initially trained in English are frequently fine-tuned for other languages, such as the CLIP model, which can be expanded with more data to enhance capabilities but can add new biases. The CAPIVARA, a CLIP-based model adapted to Portuguese, has shown strong performance in zero-shot tasks. In this paper, we evaluate four different types of discriminatory practices within visual-language models and introduce FairPIVARA, a method to reduce them by removing the most affected dimensions of feature embeddings. The application of FairPIVARA has led to a significant reduction of up to 98% in observed biases while promoting a more balanced word distribution within the model. Our model and code are available at: https://github.com/hiaac-nlp/FairPIVARA.