2024 PhD Recruitment

Join the 2024 cohort of Pioneer Centre for AI PhD students – we expect 10+ new students to start in the fall, across the centre’s partner universities. Each student will be employed at one of the partner universities AND become a part of the Pioneer Centre for AI. PhD students at the Pioneer Centre for AI have extraordinary access to computing resources, international researchers across many disciplines within computer sciences and other academic areas, as well as courses and events at the centre, and meaningful collaboration with industry, the public sector, and the start-up ecosystem.

You can read more about the Master’s and PhD study in Denmark here>>. 

To apply for one or more of the positions, follow the link after the individual project description. Note that the application deadlines vary in March and April 2024.

Here are the positions:

PhD in NLP: Factual Text Generation

While recent large language models demonstrate surprising fluency and predictive capabilities in their generated text, they have been demonstrated to generate factual inaccuracies even when they have encoded truthful information. This limits their utility and safety in real world scenarios where guarantees of factuality are needed. To address this, the project will explore methods for improving the factuality of text generation with respect to both objective real-world facts and provided source documents.

We are looking for candidates with a background in computer science, machine learning, natural language processing, computational social science, or similar. The candidate should have an interest in automatic text generation and fact checking. They should also have an interest in interdisciplinary research endeavors, including at the Pioneer Center for AI. Early research experience, especially with empirical research methods, or relevant industry experience, will be a bonus. 

The principal supervisor is Professor Isabelle Augenstein (augenstein@di.ku.dk) and the co-supervisor is Dustin Wright (dw@di.ku.dk).

April 1 deadline. For more information and to apply, click here. 

 

Task-Oriented Performance Metrics for Causal Machine Learning

Machine learning has shown that research on difficult problems can progress  remarkably quickly if clear performance metrics are available. Causal ML and causal structure learning, however, lack a universally accepted performance metric in part because graphical models can be used for many distinct tasks and simply counting edge disagreements between graphs does not reflect how similar graphs are when used for one of these causal reasoning tasks. Recently, distances for two types of graphs, directed acyclic graphs (DAGs) and completed partially directed acyclic graphs (CPDAGs), that compare graphs for the task of causal inference (i.e., using the graph to make predictions about the effects of interventions) have been developed. However, DAGs and CPDAGs are causal graphical models that we arguably will never have access to for any real-world problem, since they do not allow for unobserved variables.

The starting point of this project is the need for a performance metric that reflects similarity of causal graphs when used for a causal reasoning task, while it aims to develop metrics for the more general and arguably practically relevant class of maximal ancestral causal graphs (MAGs) that allow for hidden variables and may be learned from data. The development of a meaningful task-oriented graph metric between MAGs is important as it can enable and accelerate the development of ML/DL solutions – “to find good solutions for a task, we require a clear performance metric for candidate solutions to the task”.

The main focus of the project will be tailored to the applicant’s profile and interests and discussed at a later stage of the interview process.

The project is jointly supervised by Sebastian Weichwald (Assistant Professor, University of Copenhagen, Department  of Mathematical Sciences) and Leonard Henckel (Lecturer, University College Dublin School of Mathematics and Statistics).

April 1 deadline. Click here for more info and to apply.

 

Linguistic competence of language models in prompt interpretation and text generation

Recent generative systems based on pre-trained language models are remarkably fluent when generating even relatively exotic kinds of text, such as lymericks or texts in early middle English. At the same time, they remain sensitive to slight variation in the wording of the prompts. The proposed project will investigate this difference in competence for different linguistic phenomena when the model interprets its instructions (prompts), and generates text in response to prompts.

The specific focus is negotiable, it could be syntactic constructions, variations in lexical semantics, text registers etc. Model competence will be assessed with respect to analysis of its pre-training data (hence, an open-source model will be used). The ideal candidate would have a strong background in computational linguistics, as well as core skills in programming in Python and machine learning.

This position will be at the IT University of Copenhagen’s NLP group, under the supervision of Associate Professor Anna Rogers and Associate Professor Rob van der Goot. 

April 15 deadline. Click here to for more info and to apply.

 

Extended Reality for Behaviour Change

Extended reality (XR), including Virtual and Augmented Reality, is expected to significantly shape how we interact with computers in the future. XR technology combines advanced sensing and display capabilities, allowing us to display digital information in our physical surroundings aligned with environmental characteristics and our personal preferences.

This PhD project will investigate how to use these advanced sensing and display capabilities to help users better understand and potentially change their behaviour in their everyday lives (e.g., to help them integrate positive behaviours like physical activity into their daily lives). Relevant questions include, but are not limited to:

• How can XR support users to enhance their everyday lives through behaviour change?
• What methods and theories are useful for initiating positive behaviour change with XR?
• How should information be presented in XR to support positive behaviour change?

The PhD student will have much freedom to shape the focus of the work and the research questions. The project will use methods from human-computer interaction, including implementing and empirically evaluating XR prototypes with quantitative and qualitative methods. The project requires knowledge and interest in human-computer interaction and extended reality.

The principal supervisor is Professor Kasper Hornbæk (kash@di.ku.dk); the project is led by Assistant Professor, Teresa Hirzle (tehi@di.ku.dk). The candidate will be a member of the Pioneer Centre for Artificial Intelligence and the Human-Centered Computing research section at the University of Copenhagen’s Department of Computer Science.

March 31 deadline. Click here for more info and to apply.

 

Human-Computer Interaction in Extended Reality (XR)

Virtual (VR), Augmented (AR) and Extended Reality (XR) is increasing in popularity as a paradigm for immersive experiences.

This project will conduct foundational Human-Computer Interaction (HCI) research to advance the state-of-the-art of XR interfaces. This involves, but is not restricted to, the design, implementation, and evaluation of novel interaction concepts and systems, to better understand and shape the future of XR collaboration. The project can include, but is not limited to: 

• Adaptive XR: XR technology provides a wealth of contextual information that can be used to predict user action, and adapt the UI to the users.
• Multimodal interaction: Multimodal XR with e.g. hand, face, and eye tracking offer rich opportunities close to real-world activities, but also provide novel ways to design 3D interfaces that harness the advantages of XR tracking technology.
• Mixed physical-real interaction: Interactions can be set in different physical locations/rooms/places, IoT environments, be co-located or remote, stationary of mobile on-the-go, which raises the question how to design interaction that works across contexts.

The PhD project will take place at Aarhus University under the supervision of Professor Hans Gellersen and Assistant Professor Ken Pfeuffer.

May 1 deadline. Click here for more info and to apply.

 

Development of new node attribute measures and applications to combat financial crimes

Graph Neural Networks (GNNs) allow to apply the latest advancement in machine learning to data in complex form, that must be represented as a complex network. All GNNs techniques developed so far treat the graph as the data and allow to analyze it, performing classical network science tasks – community discovery, link prediction, etc.

However, networks can also be used to map the dimension of analysis of a problem, meaning they are not the data we are interested in analyzing, but they describe the space the data lives in. For instance, we can consider a network of financial transactions, where each node represents an economic entity and edges represent flow of capital between entities. A suspicious transaction report can be considered an observation holding a subset of transactions between entities in the graph. One could be interested not in analyzing the entities (the nodes of the graph) but instead the suspicious transaction reports: how far are two reports in this network? Can we find clusters of reports? Can some of these clusters be classified into categories of fraudulent activity?

The general aim of this PhD is to expand the state of the art of GNNs by taking this new perspective of complex networks not as data, but as maps of a complex space of interconnected dimensions. This involves progress in a number of action areas:

• Develop new basic statistical measures that are meaningful in a complex space defined by a graph. We already have a notion of Euclidean distance and Pearson correlation in a graph. We need to develop concepts such as network entropy, network cosine similarity, and more.
• Develop current methods to relax their assumptions. Currently, distances between node attributes can only be calculated for dense attributes (which have non-zero values for most nodes in the network) and networks with a single connected components. We want to relax these assumptions to increase the applicability to real world scenarios.
• Use these new statistical measures to develop new ways to perform machine learning tasks such as data clustering, creating a new class of GNN techniques.
• Apply these new GNN techniques to networks of financial transactions, with the aim of identifying potential illegal behaviors such as tax evasion, fraud and money laundering. 

The PhD project will take place at the IT University of Copenhagen under the supervision of Associate Professor Michele Coscia. The student will become a part of an award-winning lab, NERDS, and the Pioneer Centre for AI. 

March 22 deadline. Click here for more information and to apply.

 

Rigorous uncertainty estimation

For many critical applications of machine learning (ML), accurate uncertainty estimates are required. We need methods that give us uncertainty estimates for state-of-the-art ML models (e.g., deep neural networks), are well calibrated, and come with theoretical guarantees.

This project is about developing such methods. Possible theoretical approaches include PAC-Bayesian analysis (in particular for unbounded loss functions) and conformal prediction. Interesting directions are the analysis of quantile regression, Bayesian modeling and ensemble predictions.

Goals include deriving prediction intervals for non-linear regression models under heteroskedastic noise and the application of novel uncertainty estimation techniques to ecosystem/climate modeling. The main focus can be tailored to the applicant’s profile and interests and discussed at a later stage of the interview process.

The project will be jointly supervised by Christian Igel (Professor, University of Copenhagen, Department of Computer Science), email: igel@di.ku.dk and another researcher in the Pioneer Centre for AI’s Learning and Optimization Collaboratory, depending on the final PhD project description.

March 22 deadline. Click here for more info and to apply.

 

Automated detection of facial expressions of pain and stress in horses using computer vision

The Visual Analysis and Perception Lab at Aalborg University invites applications for a fully funded PhD Fellow working on novel algorithms for pain and stress detection in horses.

Recognition of pain and stress in animals is important to protect the well-being of animals. Existing work shows that pain and stress in horses may be manually recognized from the co-occurrence of certain facial activities, coded as Action Units (AU) in a facial action coding system. Based on existing datasets, this PhD project will strive to automate this process. First, the focus will be on research and development of novel computer vision methods capable of robustly detecting AUs. Next, the focus will be on novel AI methods for classifying normal vs pain vs stress faces. Lastly, head pose detection of horses in videos moving freely in a box stable will be the focus, such that we are able to identify the video segments with a near-frontal view of the head. This will allow us to validate the automated pain and stress classification algorithm on videos of horses under free-living conditions.

The candidate will be a member of the Visual Analysis and Perception Lab at Aalborg University where 25+ international researchers are working on a broad range of fundamental and applied research topics within AI and computer vision.

This project is supervised by Associate Professor Rikke Gade, rg@create.aau.dk at the Visual Analysis and Perception Lab at Aalborg University.

May 1 deadline. Click here for more information and to apply.

 

Underwater computer vision for environmental monitoring

The Visual Analysis and Perception Lab at Aalborg University invite applications for a fully funded PhD Fellow working on novel algorithms for under water computer vision.

Our marine environments are under pressure from rising temperatures, eutrophication, fishing activities, and more. This leads to a decline in biodiversity, destruction of habitats, and collapse of fish stocks, which can have huge negative societal and environmental implications. Addressing these challenges requires comprehensive and sustainable strategies that must be based on solid scientific research. However, traditional manual data gathering, and analysis tools are not scalable or sufficient to provide in-depth insights across vast underwater areas and across extended periods of time. Therefore, innovative approaches and advanced technologies are imperative to efficiently enhance our understanding and management of marine ecosystems. The PhD project will contribute to our overall effort that seeks to leverage computer vision and artificial intelligence to develop novel sensing and perception systems capable of providing unprecedented insights into the hidden world beneath the water’s surface. The goal is together with marine biologists to push the boundaries of our current understanding of underwater environments, enabling more accurate and comprehensive monitoring and conservation efforts.

In this PhD project we are particularly interested in algorithms for long term monitoring. This involves novel algorithms for detection of different types of marine organisms. The algorithms should be robust with respect to changes in turbidity and illumination, and also strive towards re-identification of the same individual across time. The team already has developed several fully annotated underwater datasets from where the PhD project can start, and more data will be collected and annotated together with our marine biologist collaborators. This includes data captured in advanced marine labs where conditions can be simulated as well as data captured in the wild using underwater mounted sensors and drones.

The PhD student will be part of a team focusing on a multitude of long term research objectives in relation to underwater computer vision for environmental monitoring. Flexibility regarding the specific research questions to focus on is an option. The candidate will be a member of the Pioneer Centre for AI, a 5-university Danish research endeavor, and physically sitting in the Visual Analysis and Perception Lab at Aalborg University. Both the center and lab are highly international and well-funded, working on a broad range of fundamental and applied research topics within AI and computer vision.

Supervisor: Professor Thomas Moeslund. 

May 1 deadline. Click here for more information and to apply.