Linköping University is now looking for a PhD student in machine learning with a focus on generative modeling and data-centric strategies for data-efficient machine learning with considerations for fairness and privacy aspects.
About the project
Machine learning, and in particular deep learning, requires large amounts of data and energy resources for training. At the same time, there are significant challenges in addressing dataset bias and privacy concerns, especially in applications that deal with sensitive data, such as medical diagnosis. The focus of this PhD project is to develop methods for reducing the dataset size without significantly affecting the performance of a model trained on that data, to promote efficient optimization and reduce the computational demands, while at the same time addressing fairness and privacy aspects. The goal is to promote resource-efficient and trustworthy machine learning in a joint framework.
In your work, you will explore generative modeling for creating synthetic representative datapoints with a high training value, while considering dataset bias and making sure that sensitive information is not leaked from the real dataset. You will work with different types of datasets (from low-dimensional point sets to high-dimensional image data) and targeting different types of applications (e.g., medical imaging). The work will be both theoretically oriented, as well as focused on implementation of experiments with machine learning algorithms for empirical testing.
The project will be conducted as a collaboration between the Division for Media and Information Technology (MIT) at the Department of Science and Technology at Campus Norrköping (principal supervisor Gabriel Eilertsen) and the Division of Statistics and Machine Learning (STIMA) at the Department of Computer and Information Science at Campus Valla in Linköping (co-supervisor Sebastian Mair).