Research map / Vision-language and multimodal / CLIP and zero-shot
LATA: Laplacian-Assisted Transductive Adaptation for Conformal Uncertainty in Medical VLMs
Behzad Bozorgtabar, Dwarikanath Mahapatra, Sudipta Roy, Muzammal Naseer, Imran Razzak, Zongyuan Ge
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add LATA graph refinement and failure-aware score to SCP (ablation) | split conformal prediction (LAC) · 9 medical tasks, alpha=0.10 | avg set size: 3.99 -> 3.07 | Improved |
| add LATA graph refinement (no ViLU) to SCP (ablation) | SCP (LAC) · alpha=0.10 | set size / CCV: Size 3.99 -> 3.14; CCV 9.96 -> 6.65 | Improved |
| add ViLU failure-aware score on top of LATA refinement (ablation) | LATA (no ViLU) · LAC alpha=0.10 | set size / CCV: Size 3.14 -> 3.07; CCV 6.65 -> 6.40 | Improved |
| add only the ViLU risk term u to SCP (ablation) | SCP (LAC) · alpha=0.10 | set size / CCV: Size 3.99 -> 3.95; CCV 9.96 -> 9.15 | Mixed |
| add only image-conditioned alpha term to SCP (ablation) | SCP (LAC) · alpha=0.10 | set size / CCV: Size 3.99 -> 3.60; CCV 9.96 -> 8.85 | Mixed |
| use LATA instead of SCA-T | transductive conformal · APS alpha=0.10 | set size / CCV / ACA: Size 3.35 -> 2.95; CCV 7.18 -> 6.32; ACA 55.2 -> 57.1 | Improved |
| turn on calibration-marginal prior (beta=0.2, LATA-LI) (ablation) | LATA-LF · APS alpha=0.10 | coverage / size: coverage 0.900 -> 0.910; size 2.95 -> 3.03 | Mixed |
| fit linear probe on calibration labels then SCP on same split (Adapt+SCP) | SCP · APS alpha=0.10 | coverage: 0.900 -> 0.858 | Hurt |
| use LATA instead of TransCLIP/TIM | transductive solvers · LAC alpha=0.10 | coverage: TransCLIP 0.726 vs LATA-LF 0.900 | Improved |
| increase window size W from 64 to 512 (ablation) | LATA · LAC alpha=0.10 | set size: ~3.12 -> ~3.06 | No clear effect |
| increase calibration shots K from 4 to 16 (ablation) | LATA · LAC alpha=0.10 | set size: 3.18 -> 3.08 | Improved |
| use LATA refinement instead of SCA-T/TIM for compute | transductive solvers · LAC alpha=0.10 | runtime (s): 1.04 -> 0.05 (vs SCA-T) | Improved |
Closest papers by meaning
- Conformal Prediction for Zero-Shot Models (CVPR 2025 · similarity 0.87)
- Scaling Full Conformal Image Classifiers (NeurIPS 2026 · similarity 0.87)
- How Do Medical MLLMs Fail? A Study on Visual Grounding in Medical Images (ICLR 2026 · similarity 0.86)
- An Empirical Study Into What Matters for Calibrating Vision-Language Models (ICML 2024 · similarity 0.85)
- Free on the Fly: Enhancing Flexibility in Test-Time Adaptation with Online EM (CVPR 2025 · similarity 0.85)
- Uncertainty Estimation for Pretrained Medical Image Registration Models via Transformation Equivariance (NeurIPS 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
