Digital Systems

Digitalization opens many doors to new knowledge and gives us new ways to understand and work with both old and new problems. Success requires collaboration across various disciplines which ensure the whole value chain of data flow, from collection and quality control to how to reap the benefits of the data and how to interact with them. Below follows a brief overview over important themes for our research group.

Data Processing

Effective data processing is particularly important when the resources are limited. An example of such a problem is for example surveys done in harsh environments with limited access to processing and network capacity. Good quality data processing is also the corner stone for further analysis with for example machine learning. Insight starts with good data.

Visualization

Visualization of data influences what data gathers our attention and what decisions we make. Effective visualization can make difficult problems manageable, nudge us in the right direction and make our everyday life easier. Visualization is therefore an important part of digitalization processes. For projects where one analyzes different datasets across competence fields, we have developed an advanced visualization software named Enlighten which facilitates the work. Below you may see a video from the project EPOS-N, European Plate Observation System – Norway, which shows the software in action.

Visualization of data can be a great tool for decision support. If visualizations are to be used as a support tool in work processes one needs to take into account information needs, backgrounds and working goals that are part of the individual’s context, in addition to what type of data are available for visualization. A good way to achieve useful visualizations are through user driven design. An example of a user driven design process we have been working with is the visualization of patient data for clinicians in an online tool for cognitive behavioral therapy. The visualization helps to identify which patients need closer follow up and how to prioritize work.

Artificial Intelligence and Machine Learning

Artificial intelligence and machine learning make it easier to gain new insight where classical data analysis struggle with the amount of data. Machine learning finds patterns and connections in big and complex data sets. Different data sets and application areas demand different types of machine learning algorithms, and it is important to know the limitations that govern their uses and the accuracy which can be obtained. We work cross disciplinary and in close collaboration with the users to ensure the best possible results and with the highest utility values. Responsible AI is particularly important when working with persons and personal data, and the implications of choices made in the development of the algorithm must be evaluated from different points of view. To explain how the machine learning algorithm arrives at the results increases trust in the process and can diminish user resistance – so called explainable AI (XAI). We work with different markets and application areas, and our project portfolio includes for example projects within transport, oil and gas, aquaculture, healthcare and renewable energy.

An example of use of machine learning within aquaculture is a project we have developed for a customer for counting fish lice in a salmon farm. Here we have used deep learning models, where one uses several layers of neural networks to localize salmon and detect and count fish lice by means of video imaging. Quality control of the prevalence of fish lice is important for evaluating which control measures should be put into motion when.

Deep learning models are also used in the analysis of time signals, for instance in predictions and prognoses. A commonly used model is recurrent neural networks (RNN). In one of our projects we have developed a coder-decoder architecture based on RNN for wind speed predictions, which shows promising results for short term predictions. Accurate predictions can for instance increase the profitability of wind farms when using accurate wind predictions in combination with advanced control systems.
Other advanced deep learning models like Transformer have also been successfully applied in our research and projects. We have for instance proposed an advanced long-short term convolutional Transformer (LSCT) for video quality assessment, which demonstrates outstanding performance against state-of-the-art methods.

Data analysis and decision support systems

Software developed in close collaboration with the user can give easier access to the data analysis needed and with an information flow that suits the working situation best. We have a long experience with developing advanced analysis- and decision support systems, like for instance SARA and LSSS.

SARA has been operational since the year 2000 and is used by HRS (Norwegian Rescue Services), Radio Medico and the Norwegian Coastal Radio in order to coordinate rescue missions. The main function of the software is to coordinate and present all relevant information about real and potential emergency situations, ensuring ease of communication between the parties. SARA is short for Search And Rescue Application.

LSSS has been in use since 2007 and builds on a long and constructive collaboration with IMR (Institute of Marine Research). The software is particularly useful for researchers and is used to interpret acoustic data from the ocean. These data are then used to estimate the size of fish populations which is a very important factor in the decision of fishing quotas. LSSS is short for Large Scale Survey System.