The Norwegian Centre for Knowledge-driven Machine Learning

About Integreat

Integreat is a Norwegian Centre of Excellence (SFF), funded by the Research Council of Norway. The centre aims is to develop knowledge-driven machine learning, where data are combined with statistics, logic and domain knowledge to make artificial intelligence (AI) more precise, ethical and sustainable.

Machine learning lies at the core oe3sdf artificial intelligence and is widely used in a range of critical systems across sectors. At the same time, modern machine learning raises a number of challenges related to explainability, fairness and the handling of uncertainty. Integreat develops new methodological frameworks designed to address these challenges.

NR contributes primarily thorugh methodological development in explainability, anomaly detection and fair decision-making under certainty. Our work is characterised by a strong foundation in statistical modelling combined with modern machine learning techniques.

Explainable artificial intelligence

Many machine learning models function as “black boxes,” making it difficult to understand why a model produces a particular result. Deep learning is highly effective at identifying complex patterns in data, but it often provides limited insight into how decisions are made. By contrast, statistical models are typically more interpretable but may be less flexible when dealing with highly complex data structures.

At NR, we develop methods that combine the strengths of these approaches. We place particular emphasis on modelling the interactions between variables and on developing explanations that are both methodologically robust and practically applicable. The aim is to build models that not only deliver accurate predictions, but can also be justified and verified.

Anomaly detection

Many machine learning models are built on the assumption that data remain relatively stable over time. When the underlying data change, a model’s precision and stability may deteriorate. Ensuring reliable operations therefore requires the ability to detect anomalies quickly and accurately.

We develop statistical and machine learning-based methods for detecting anomalies in data streams in real time. Our work focuses on reliable and adaptable algorithms that can handle evolving data patterns while providing clear criteria for when something actually deviates.

The aim is for these methods to be applied in real-world settings in order to monitor AI-based systems and strengthen and quality-assure their performance.

Fairness and uncertainty

When machine learning models are used to make decisions that affect people, fairness must be explicitly addressed. Without a deliberate methodical approach, existing biases may be perpetuated or even amplified.

We develop methods to detect and quantify bias in models, for example by analysing how changes in protected characteristics, such as gender or age, influence outcomes. We also examine how biases evolve over time and what long-term consequences different interventions may have.

Another strand of our research focuses on how fair decisions can be made when the preferences of the affected parties aren’t fully known. In practice, this involves developing methods that prevent certain groups from being systematically disadvantaged, even when decisions are made on the basis of incomplete information.

To learn more about Integreat
and our role in the centre, get in touch.

Explore our focus areas