4 Funded PhD Positions in Fellowships in the Pioneer Centre for Artificial Intelligence’s collaboratories, University of Copenhagen

The Pioneer Centre for AI invites applicants for 4 PhD fellowships with start dates in September 2022 or as soon as possible thereafter. The PhD stipends are funded or co-funded by the Pioneer Centre for Artificial Intelligence’s collaboratories: Extended Reality, Fine-grained Analysis, Networks & Graphs, and Causality & Explainability in the Department of Computer Science and Department of Mathematical Sciences.

The Pioneer Centre for AI is located at the University of Copenhagen, with partners at Aarhus University, Aalborg University, Technical University of Denmark, and the IT University of Copenhagen. The Pioneer Centre for Artificial Intelligence is a multi-university research centre led by Professor Serge Belongie. Interdisciplinary and at the forefront of fundamental AI research, the centre focuses on fundamental AI research, and within an interdisciplinary framework, develops platforms, methods, and practices addressing society’s greatest challenges.

These PhD positions are offered at the University of Copenhagen (UCPH). UCPH was founded in 1479 and is the oldest and largest university in Denmark. It is often ranked as the best university in Scandinavia and consistently as one of the top places in Europe. Within computer science, it is ranked 2nd in the European Union (post-Brexit) by Shanghai Ranking.

The projects in the Funded PhD Fellowships

1. Causal Abstraction and Fairness: This project explores how to use abstract causal knowledge as weak supervision signal for fair machine learning (ML). Many definitions of fairness build on causal models. Yet, we cannot readily use these notions in practice if we have only partial causal knowledge instead of the full causal model. The full causal model is arguably often unattainable. Further, our prior knowledge is often not only partial but comes in abstract form of high-level properties instead of exact causal constraints on the model variables; for example, we may only know that a recruitment process suffers from gender bias, but not the exact causal pathways for all aspects of the application material (name, picture, distinct CV timelines) and the interview process (perceived competence, strong appearance). To incorporate such abstract causal knowledge, we need a theory of causal abstraction. Recent frameworks of causal abstraction, however, build on exact interventions, which are not pragmatically useful when applicants cannot possibly intervene on certain causal factors. Instead, we will characterise the correspondence between invariance of causal mechanisms (cf. autonomy, modularity) on a detailed level and (novel) regularity properties on the abstracted causal mechanisms. We investigate whether fairness can be defined in terms of these high-level regularities and without the notion of intervention or access to a full fine-level causal model. This work may shed light on the limits of what a general definition of fairness can achieve and may have implications of applying fairness constraints in settings where causal knowledge is hard to obtain.Principal supervisor: Assistant Professor Sebastian Weichwald, Department of Mathematical Sciences, [email protected]. Co-supervisor, Professor, Jonas Peters, Department of Mathematical Sciences.
2. Moving in Virtual Reality: This PhD project is in the field of human–computer interaction and virtual reality. Virtual reality (VR) allows people to move in physically impossible, risky, or distant situations, such as to share the control of their body with a yoga teacher or jump on the moon to experience space travel. Such VR experiences depend crucially on how people’s physical movements are mapped to the ones of their virtual body. For example, a comfortable vertical position of the physical hand can be mapped to reach the virtual hand at a much higher position; or a short walking distance in a physical room can be mapped to cover a much larger virtual space. However, it is still not known how to map movements in a way that supports learning in and outside VR. In this project, we will design, evaluate, and develop interaction techniques to move in VR. The project involves collaborations with Neuroscience on motor learning. The PhD fellow will be part of the Pioneer Center for AI and the Human–Centered Computing section at the University of Copenhagen. The section aspires to be a top research environment for human-computer interaction that bridges technology, theory, and empirical studies, and offers a large and active group of fellow PhD students for peer-support and for engaging with in social activities. Thereby, this position offers an opportunity to join a globally recognized, highly ambitious, creative, collaborative, and fun group of international researchers working at the forefront of HCI, XR, and AI. Principal Supervisor: Associate Professor, Joanna Bergström, Department of Computer Science, [email protected]. Co-supervisor: Associate Professor Mark Schram Christensen, Department of Neuroscience.
3. Concept bottleneck for XAI: This project will develop and study (X)AI feedback for clinicians while they learn to perform fetal ultrasound screening scans. Ultrasound is cheap and non-invasive, and therefore widespread, but it is highly operator dependent. As the value of recorded images for downstream predictions depends strongly on the image quality, access to specialized clinicians is crucial to the quality of the performed screening. The ultimate goal of this project is to develop AI tools that train and support clinicians in obtaining a standardized, high level of image quality by providing feedback during the scan. To do so, this project will implement and evaluate a range of basic feedback models for detecting “standard ultrasound planes” that satisfy standardized quality conditions. This basic infrastructure will be developed in collaboration with other PhD students affiliated with the Pioneer center collaboratories for Causality and Explainability and Fine-Grained Analysis. These feedback models will next be transferred to the Simulation Labs (Sim-Labs) located at CAMES Rigshospitalet, where the AI feedback will be tested by clinicians while training. Technically, the project aims at establishing measures of uncertainty of features identified in the US images. Ideally the aspects that must be measured like femur length and skull width are based on clearly visible US features. Uncertainty of these features drives the feedback to the clinicians that must navigate with the US probe to make features certain and measurements with as little noise as possible. Principal supervisor: Professor Mads Nielsen, Department of Computer Science, [email protected]. Co-supervisor, Professor Aasa Feragen, Technical University of Denmark.
4. Embedding life and health: During your journey through life, you leave behind digital remnants in the health, the social, the educational, the legal system, and many more. This is a rich source of information for finding and visualizing patterns in life trajectories. Technically, we observe a number of streams of events in several channels, and a number of interactions with other individuals. The purpose of this research is to structure these as embeddings in vector spaces making visualizations, clustering, predictive analytics etc possible. Driving questions are in pharmacovigilance, in long covid tracking, in social interactions. Can we develop methods for detecting the impact on life of medications, of having had Covid, and can we quantify or even predict this? Methodologically, we will develop deep networks inspired by the foundation models in Natural Language Processing like BERT and GPT-3 by graph neural netwrks and by variational autoencoders. Data-wise the project will rely on already harvested electronic health records from Capital Region and Zealand Region of Denmark comprising 2,4 mio subjects and data from Statistics Denmark on socioeconomic factors from the Danish population. Principal supervisor: Professor Mads Nielsen, Department of Computer Science, [email protected]. Co-supervisor: Professor Sune Lehman, Technical University of Denmark.

APPLY NOW