CLIP the Bias: How Useful is Balancing Data in Multimodal Learning?
This work addresses bias mitigation in multimodal AI systems, which is crucial for fairness in applications like image classification and retrieval, though it is incremental as it builds on prior bias studies in CLIP.
The study investigated the effectiveness of data-balancing for mitigating biases in CLIP models, finding that fine-tuning reduces representation biases but has limited impact on association biases, with data balancing improving classification but sometimes harming retrieval, as shown by a 1% increase in COCO retrieval and 0.5% in ImageNet classification when using their M4 algorithm with quality filters.
We study the effectiveness of data-balancing for mitigating biases in contrastive language-image pretraining (CLIP), identifying areas of strength and limitation. First, we reaffirm prior conclusions that CLIP models can inadvertently absorb societal stereotypes. To counter this, we present a novel algorithm, called Multi-Modal Moment Matching (M4), designed to reduce both representation and association biases (i.e. in first- and second-order statistics) in multimodal data. We use M4 to conduct an in-depth analysis taking into account various factors, such as the model, representation, and data size. Our study also explores the dynamic nature of how CLIP learns and unlearns biases. In particular, we find that fine-tuning is effective in countering representation biases, though its impact diminishes for association biases. Also, data balancing has a mixed impact on quality: it tends to improve classification but can hurt retrieval. Interestingly, data and architectural improvements seem to mitigate the negative impact of data balancing on performance; e.g. applying M4 to SigLIP-B/16 with data quality filters improves COCO image-to-text retrieval @5 from 86% (without data balancing) to 87% and ImageNet 0-shot classification from 77% to 77.5%! Finally, we conclude with recommendations for improving the efficacy of data balancing in multimodal systems.