This is a final-year project group (CO425) project. The research focused on fine-tuning strategies for CLIP-based Vision Language Models (VLMs) to improve generalization in unseen medical imaging domains. I introduced a novel evolutionary algorithm based prompting strategy, which utilizes LLMs as optimizers and biomedical knowledge extractors for prompt optimization for Biomedical VLMs. The project also explored common CLIP finetuning techniques for out-of-domain performance in medical imaging.
Out-of-Domain Generalization in Medical Imaging via VLMs
January 2025
university ai research machine-learning
Research on fine-tuning strategies for CLIP-based Vision Language Models to improve generalization in unseen medical imaging domains.