Hidden-uncertainty Assessment via Non-verbalized Signatures for Medical QA - Models & Results

Description: This dataset contains the experimental results and trained models for the paper "Hidden-uncertainty Assessment via Non-verbalized Signatures for Medical QA". Contents - results/ - Per-model experimental results including: - Trained classifiers (Random Forest, Logistic Regression, MLP) for error prediction - Cross-validation predictions and feature importance scores - Results across 10 medical QA benchmarks - results_common/ - Cross-model transfer experiments: - Universal feature models trained with Leave-One-Dataset-Out (LODO) - Self-verbalization comparison experiments Datasets Covered MedQA, MedXpertQA-R, MedXpertQA-U, AfriMedQA, MedBullets, MedExQA, MedMCQA, MMLU-Pro, MMLU (medical), PubMedQA Models Analyzed - DeepSeek-R1-Distill-Qwen-32B - Qwen3-32B - Olmo-3-32B-Think (AllenAI) - gpt-oss-120b - gpt-oss-120b-high