Work

Fa-DPO: Faithfulness-Aware Direct Preference Optimization with risk-adaptive preference learning for medical question answering