Evo-PI framework boosts medical visual QA accuracy

According to Sciencecast, researchers introduced the Evo-PI learning framework, which lets supervision principles evolve during training, boosting accuracy of multimodal AI models on medical visual question answering tasks. The framework treats reasoning principles as language-based signals that can be generated, evaluated and iteratively refined. In benchmark tests, it lifted accuracy on medical visual question answering by up to 24.

6 percent. The team also released the code publicly to support further research. The authors noted that adaptive supervision can enhance multimodal performance in specialized domains.