When Choices Become Priors: Contrastive Decoding for Scientific Figure Multiple-Choice QA

1. 1Identify the bias in scientific figure MCQA where answer choices act as priors
2. 2Develop a contrastive decoding approach to utilize answer choices as priors
3. 3Implement the contrastive decoding method to mitigate the bias and improve model accuracy
4. 4Evaluate the effectiveness of the approach on a dataset of scientific figures and multiple-choice questions

AI researchers and engineers working on multimodal models can benefit from this approach to improve the accuracy of their models, especially in scientific figure multiple-choice question answering tasks

💡 Utilizing answer choices as priors can help improve the accuracy of multimodal models in scientific figure multiple-choice question answering

💡 Mitigate bias in scientific figure MCQA with contrastive decoding!