CoDA: Exploring Chain-of-Distribution Attacks and Post-Hoc Token-Space Repair for Medical Vision-Language Models

1. 1Identify potential chain-of-distribution attacks on medical vision-language models
2. 2Develop post-hoc token-space repair methods to mitigate these attacks
3. 3Evaluate the effectiveness of these repair methods under various clinical workflows
4. 4Integrate these methods into existing radiology pipelines and multimodal assistants

Researchers and developers working on medical vision-language models can benefit from this study to improve the reliability of their models under real clinical workflows, and ML engineers can apply these findings to develop more robust models

💡 Chain-of-distribution attacks can compromise the reliability of medical vision-language models, and post-hoc token-space repair can mitigate these attacks

🚨 CoDA explores attacks on medical vision-language models & post-hoc repair methods 🚨