Uncertainty quantification of missing data
Dive into an experimental study to discover the best methods for quantifying uncertainty when filling in missing values for time series data and how this affects the uncertainty quantification in downstream classification tasks. Potential applications are health care, sport analysis, or lifelogging (the application of your choice).
The aim of the project is to discover the best methods for quantifying uncertainty when filling in missing values for time series data, and how this affects the uncertainty quantification in downstream classification tasks, with application in health care, sport analysis, or lifelogging. Investigate the challenges posed by missing values in time series data and their potential impact on the reliability of downstream analyses. By examining the application of various uncertainty quantification techniques in different domains, this study aims to provide valuable insights into enhancing the accuracy and robustness of classification tasks in the chosen application areas.
Goal
Discover the best methods for quantifying uncertainty when filling missing values for time series data, and how this affects the uncertainty quantification in downstream classification tasks
Learning outcome
- Missing data analysis and how to apply it to e.g. sport or lifelogging data
- Machine learning and deep learning
- Building proper research methodology to handle missing data and evaluate the performance of different techniques
- Writing a scientific paper and how to publish it
Qualifications
- Hard working, motivated
- Fluent in Python programming
- Interested in learning (the rest can be learned during the thesis work)
Supervisors
- Thu Thi Nguyen
- Luis Lopez Ramos
- Hugo Hammer
References
- A Python Toolbox for Machine Learning on Partially-Observed Time Series
- Supervised Machine Learning for Science, chapter 13 Uncertainty
