Multimodal Vision Research Laboratory

MVRL

Research Area: Uncertainty Estimation

When should a model abstain or flag low confidence—and how do we calibrate neural predictors for deployment? We study uncertainty estimation and calibration for vision models, with emphasis on reliable confidence for safety-sensitive settings. Recent work includes improved trainable calibration for neural networks, decision-making criteria consistency via input sensitivity, and calibration approaches applied to medical imaging classification. Uncertainty methods support trustworthy use of MVRL models in healthcare and geospatial pipelines where errors carry real-world cost.

View All Publications →

All Publications

  1. Xing E, Xing X, Liu L, Jacobs N, Qu Y, Liang G. 2022. Neural Network Decision-Making Criteria Consistency Analysis via Inputs Sensitivity. In: International Conference on Pattern Recognition (ICPR 2022). DOI: 10.1109/ICPR56361.2022.9956394.
    doi bibtex Uncertainty
  2. Thumbnail for Improved Trainable Calibration Method for Neural Networks
    Liang G, Zhang Y, Wang X, Jacobs N. 2020. Improved Trainable Calibration Method for Neural Networks. In: British Machine Vision Conference (BMVC).
    website paper bibtex Medical Uncertainty
  3. Liang G, Zhang Y, Jacobs N. 2020. Neural Network Calibration for Medical Imaging Classification Using DCA Regularization. In: ICML 2020 workshop on Uncertainty and Robustness in Deep Learning (UDL).
    paper bibtex Medical Uncertainty