Uncertainty quantification for data driven clinical decision making

Title Uncertainty quantification for data driven clinical decision making
Summary The student will build upon the field of evidential deep learning to identify and understand when the model says 'I don't know'
Keywords Evidential deep learning, uncertainty quantification, electronic health records
TimeFrame 2023-2024
References Sensoy, Murat, Lance Kaplan, and Melih Kandemir. "Evidential deep learning to quantify classification uncertainty." arXiv preprint arXiv:1806.01768 (2018).

Amini, Alexander, et al. "Deep evidential regression." arXiv preprint arXiv:1910.02600 (2019).

Prerequisites Statistics; Neural Networks; Programming (Python or Matlab)
Supervisor Awais Ashfaq, Slawomir Nowaczyk
Level Master
Status Open

Neural networks are increasingly being used in many safety critical decision processes, thereby necessitating reliable uncertainty estimates along with predictions. Healthcare is no different and reliable and accurate predictions carry great importance, since they may contribute to an incorrect clinical decision risking a human life in addition to severe ethical and financial costs.

Data for the project comes from MIMIC-III (an open source EHR database)

The student will go through the state-of-the-art in Bayesian and Evidential Learning to quantify prediction uncertainties. Focus will be on identifying and understanding when the model says "I don't know".

For questions contact Awais Ashfaq (awais.ashfaq@hh.se)