Uncertainty estimation is vital to ensure trustworthy predictions from ML models. In particular, we desire that models ‘know what they don’t know’. The lab has developed a variety of uncertainty estimation methods which can be found below.
Additionally, we’ve developed data-centric uncertainty estimation methods, where we ask how can your current data be re-used but also any additional data (such as unlabelled data) be used to improve the efficiency of the uncertainty estimates
The lab’s work on uncertainty estimation:
Research Pillar: https://www.vanderschaar-lab.com/uncertainty-quantification/
Read more:
Self-supervised Conformal Prediction – how unlabelled data can be used to improve conformal prediction: https://arxiv.org/abs/2302.12238
