Economic boost for digital research at University of Copenhagen

In total, six researchers at DIKU have just received a grant from Independent Research Fund Denmark (DFF) for their research projects, that will contribute in various ways to the development in the digital field in Denmark.

- Digitalisation is a strategically important area, not only at DIKU but at the entire University of Copenhagen. Several of the projects extend across scientific fields both within and outside University of Copenhagen and thus, show the broadness in the digital area with applications in areas such as climate, health, law and communication. It's digitalisation that makes sense in practice - not digitalisation for the sake of digitalisation - and we're very happy that we now have the opportunity to contribute to this even more, says Mads Nielsen, professor and head of department at DIKU.

All the projects will be initiated in the period January - April 2020.

Monitoring of the Earth's surface will give insights into climate change

Today, huge amounts of image data are collected from satellites that can help map and monitor forest areas, agriculture, urban extensions, changes in water volume, burned areas and much more. This data is crucial to reduce the effect of global warming, as it can provide us with information on ecosystem changes that could have major impacts on the climate. However, at present, the artificial intelligence (AI) models used to analyse this data can take weeks, months and even years due to the large amount of data.

In the new research project Monitoring Changes in Big Data via Massively-Parallel Artificial Intelligence, assistant professor Fabian Gieseke in collaboration with two other researchers from DIKU as well as the Department of Geosciences and Natural Resource Management, the Danish Meteorological Institute (DMI) and Wageningen University in the Netherlands will develop effective implementations of GPUs (a visual processing device used, among other things, in computer games) in AI models that can analyse this data more efficiently than today.

The project has received DKK 2.876.381

More effective sleep analysis via deep learning 

An increasing number of Danes suffer from sleep disorders, which may correlate with diseases such as depression, schizophrenia, Parkinson's and Alzheimer's. Today, approx. 3000 sleep studies are conducted annually by the Danish Center for Sleep Medicine, mostly manually, which require several hours of work by the clinical staff. Deep learning models that can perform sleep analysis exist, but are not widely used in Danish clinics as the models are complex and difficult to interpret and implement in clinical practice.

In the project U-Sleep (with reference to the neural network U-net), professor Christian Igel will develop a sleep analysis system that can analyse sleep with higher than human performance. This system is based on a different class of deep learning models that are more robust in clinical use than current ones. These models have not previously been used in this field, but have proven to be extremely successful in, among other things, medical image analysis.

The project has received 2.616.761 kr.

Better personal data in public and private digitalisation

The increasing digitisation of public and private institutions, that process personal data from various sources to provide personalised services tailored to the context, places new demands on the programming languages and systems used to write and execute the programs.

In the project PAPRiCaS: Programming technology foundations for Accountability, Privacy-by-design & Robustness in Context-aware Systems, professor Thomas Hildebrandt will establish a collaboration with leading European research environments to contribute to new knowledge and research capacity in the development of programming languages and systems, that in the future can ensure the correct handling of personal data, documentable responsibility and robustness of IT systems consisting of many different parts.

The project has received 5.900.522 kr.

The robots of the future are becoming more human

A new type of robots have seen the light of day; flexible and soft robots that can work with humans without harming them. These robots have much greater freedom of movement, just like humans, unlike the industrial, hard robots. This is a great advantage in many applications - but it also makes it much more difficult to design and program them, and no technology is available for this yet.

In the new project Reward Modeling for Soft Robotics AI, associate professor Kenny Erleben, who heads the world's leading research group in robotic simulation methods, will in collaboration with IT University of Copenhagen, NVIDIA and McGill University in Canada use artificial intelligence (AI) to develop technology for this type of robots. Specifically, they will invent a learning model that tells the computer how to reward a task being performed better - so-called reinforcement learning. In addition, the shape and function of the robot will also be improved via virtual simulation.

The project has received 2.582.150 kr.

Detecting gender bias on social media

When we write and speak, we choose, consciously or unconsciously, which words we will use among a wide variety of words. These choices can be greatly influenced by our gender as well as tell something about what attitude we have toward other people.

In a pilot study, assistant professor Isabelle Augenstein previously investigated 3.5 million English books for gender bias for adjectives and verbs. In the project Detecting Gender-Biased Language on Social Media project, she will take the next step in detecting and quantifying gender bias in connection with named entities - in any language - based on politicians on social media.

The project has received 2.875.582 kr.

Computers should predict the three-dimensional structure of proteins

Proteins have a three-dimensional structure that is fundamentally related to their biological function. But discovering this three-dimensional structure is often expensive, time-consuming, and challenging - and predicting this structure is yet an unsolved mystery for computers.

In the project Deep Probabilistic Programming for Protein Structure Prediction, associate professor Thomas Hamelryck will study the so-called protein folding problem, which is important in areas such as medicine, biotechnology and science in general. He will tackle the problem by using an emerging technology in machine learning; probabilistic programming, which allows to combine deep learning (one of the most powerful methods in machine learning) with Bayesian statistics, which allows to assign uncertainties to predictions. In addition, the project will also contribute to the development of new machine learning methods.

The project has received 5.901.438 kr.