Revealing Epistemic Uncertainty in MLLMs via Causal-Invariant Masking

arXiv:2610.02887 · cs.CV, cs.AI · Submitted 2026-10-02 · Read on arXiv

cs.CV, cs.AI

Submitted: 2026-10-02

Updated: 2026-10-02

Code: https://github.com/RUCAIBox/POPE

Project page: https://cs.stanford.edu/people

Terminology

• Epistemic Uncertainty — 1× in this paper · explained in 10 episodes

Sources

• Invariant Risk Minimization
• Qwen2.5-VL Technical Report
• A Survey of Multimodal Large Language Model from A Data-centric Perspective
• Hallucination of Multimodal Large Language Models: A Survey
• Autoregressive Language Models are Secretly Energy-Based Models: Insights into the Lookahead Capabilities of Next-Token Prediction
• Semantic Entropy Probes: Robust and Cheap Hallucination Detection in LLMs
• Large Language Models: A Survey
• MV-CoRe: Multimodal Visual-Conceptual Reasoning for Complex Visual Question Answering
• Revisiting Uncertainty Estimation and Calibration of Large Language Models
• Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs
• VL-Uncertainty: Detecting Hallucination in Large Vision-Language Model via Uncertainty Estimation
• Uncertainty-o: One Model-agnostic Framework for Unveiling Uncertainty in Large Multimodal Models
• A Survey of Large Language Models

Related papers

• Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
• AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
• Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
• MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
• TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
• A Survey on Efficient Vision-Language-Action Models