Master’s thesis - Uncertainty Quantification and Active Learning

This thesis will be carried out in conjunction with the CellFace project, which investigates the potential of computer vision and machine learning techniques for the task of automated blood cell diagnosis.

Topic

Although deep learning is a booming area of research, the translation of developments in this field to real world applications is often impeded by safety and reliability concerns. This is especially the case in domains like medicine, where relying on the wrong decision of a classifier could have very serious consequences. It has repeatedly been shown that the most popular and widely used deep learning models are unable to provide accurate estimates of how much trust should be put in their outputs. They tend to make overconfident predictions, even on data that is very different from the data used to train them. A range of uncertainty quantification methods for deep learning have been suggested in the recent literature to address this issue. The goal of the proposed thesis is to evaluate and compare the usefulness of these different uncertainty quantification approaches in the context of active learning.


The thesis will be carried out in conjunction with the CellFace project, which you can read more about here.

Requirements

  • Python coding skills
  • Prior knowledge of deep learning techniques
  • Capacity for teamwork
  • Experiences with Git

For a more in-depth introduction to the relevant research topics, see:

Guo, Chuan, Geoff Pleiss, Yu Sun, and Kilian Q. Weinberger. 2017. “On Calibration of Modern Neural Networks.” ArXiv:1706.04599 [Cs], August. http://arxiv.org/abs/1706.04599.

Siddhant, Aditya, and Zachary C. Lipton. 2018. “Deep Bayesian Active Learning for Natural Language Processing: Results of a Large-Scale Empirical Study.” ArXiv:1808.05697 [Cs, Stat], September. http://arxiv.org/abs/1808.05697.

Nguyen, Vu-Linh, Sébastien Destercke, and Eyke Hüllermeier. 2019. “Epistemic Uncertainty Sampling.” ArXiv:1909.00218 [Cs, Stat], August. http://arxiv.org/abs/1909.00218.

Supervisors

Alice Hein, M.Sc. and Stefan Röhrl, M.Sc.

Chair for Data Processing

Contact Information

Alice Hein, M.Sc.

Chair for Data Processing

TUM Department of Electrical and Computer Engineering

Technical University of Munich

 

Arcisstr. 21, 80333 Munich

Room Z942

alice.hein@tum.de