Similarity-as-Evidence: Calibrating Overconfident VLMs for Interpretable and Label-Efficient Medical Active Learning

Documenti analoghi

• TIR-Flow: Active Video Search and Reasoning with Frozen VLMs
  di: Jin, Hongbo, et al.
  Pubblicazione: (2026)
• Overconfidence and Calibration in Medical VQA: Empirical Findings and Hallucination-Aware Mitigation
  di: Byun, Ji Young, et al.
  Pubblicazione: (2026)
• Dynamic Correlation Learning and Regularization for Multi-Label Confidence Calibration
  di: Chen, Tianshui, et al.
  Pubblicazione: (2024)
• MedConcept: Unsupervised Concept Discovery for Interpretability in Medical VLMs
  di: Haque, Md Rakibul, et al.
  Pubblicazione: (2026)
• Are We Overconfident in Models and Results for Semi-Supervised 3D Medical Image Segmentation?
  di: Li, Jun, et al.
  Pubblicazione: (2026)