Back to search

SFI-Sentre for forskningsdrevet innovasjon

Visual Intelligence

Alternative title: Visuell Intelligens

Awarded: NOK 96.0 mill.

As a Center for Research-based Innovation (SFI), Visual Intelligence is propelled by the interaction between the business sector's need for new solutions in artificial intelligence (AI), public institutions' requirements for digital transformation, and the necessity for fundamental research in deep learning methodologies and image analysis to establish a foundation for innovations. Visual Intelligence serves as a hub in Norway for deep learning and artificial intelligence, recognized internationally, as reflected in the quality and scale of the center's research and innovation and in the centre's international Northern Lights Deep Learning Conference (http://nldl.org) organized in Tromsø each January. The center's core lies in close collaboration between user partners from the business and public sectors, including • Equinor • Field • GE Vingmed Ultrasound • Institute of Marine Research • Kongsberg Satellite Services • Cancer Registry of Norway • University Hospital of North Norway/Helse Nord IKT along with the research partners in artificial intelligence • UiT The Arctic University of Norway • Norwegian Computing Center • University of Oslo The center focuses on leveraging synergies and interdisciplinary expertise across medicine and health, marine science, the energy field, and Earth observation. In 2023, Visual Intelligence (VI) delivered numerous results in the form of innovations for new solutions based on methodological research in deep learning. Specific achievements related to each user partner include: • Equinor: The company has an operational system for oil exploration using AI analysis of seismic images developed by the Norwegian Computing Center prior to the start of VI. The research in VI has led to new approaches for extracting information from seismic data, now under evaluation, potentially providing new AI-based tools for subsurface analysis. VI has furthermore developed prototypes for analysis of microfossil images from subsurface drill boreholes. This is important for potential carbon capture and storage. • Field: Visual Intelligence developed prototypes using self-supervised learning to automatically detect changes in building structures over time from image data, crucial for automated mapping and potential cost savings. • GE Vingmed Ultrasound: The center conducted research towards new solutions to expand the AI-based toolkit for the company's ultrasound scanners. Several prototypes integrating knowledge of heart anatomy into deep networks were developed and are under evaluation in 2023. The prototypes may be integrated into coming versions of the ultra sound scanner. • Institute of Marine Research (HI): The center researched new types of deep learning for the analysis of acoustic image data, focusing on better utilizing unannotated data and improving uncertainty estimates in deep learning. These solutions are being evaluated by HI in their data flow from research cruises. The potential impact of automated monitoring of fish for better abundance estimation is regarded to be high. • Kongsberg Satellite Services (KSAT): KSAT tested a new prototype for ship detection using radar images from satellites based on Visual Intelligence's research in 2023. KSAT’s operational system for ship detection is based on deep learning. • Cancer Registry of Norway: Visual Intelligence developed AI methods for improved mammography screening through automatic analysis of mammogram images, being tested on the Cancer Registry's mammography screening data. Methods for evaluating the quality of mammography images exploiting interpretable AI is under development and under evaluation by radiologists. • University Hospital of North Norway (UNN)/Helse Nord IKT: The center's research led to a new method for diagnostic support in nuclear medicine PET imaging using deep learning, possibly moving towards software licensing through collaboration with Norinnova. Visual Intelligence focuses heavily on addressing fundamental challenges in deep learning common to innovation areas and operates at a high international level, delivering research at the level of a Center for Excellence. Their new methods learn from limited data, leverage context, quantify uncertainty, and are interpretable. In 2023, Visual Intelligence published in the top journals and conferences in the field of deep learning, including CVPR, ICLR, NeurIPS, etc.

The user partners of Visual Intelligence all aim at becoming more data-driven, where information extraction from digital visual data is an essential part of this and important for their value creation. They all have very complex imagery, acquired from a variety of sensors. The biggest drivers in the recent progress of AI systems for computer vision is the use of deep learning. However, there is still a long way before the full potential of deep learning is realized for applications and industries relying on more complex visual data. This is especially the case when annotated visual data are scarce and experts are needed to interpret them. Visual Intelligence aims to unlock the unused potential of deep learning methodology for extraction of knowledge from complex image data. To achieve this the centre will develop solutions for (i) learning from limited data; (ii) exploiting context and prior knowledge; (iii) estimation of confidence and uncertainties; and (iv) explainable models. This will create value for the user partners across our main innovation areas: -Medicine and health -Marine science -Energy and industry -Earth observation. Through this the centre will develop innovations that will contribute to solving important societal challenges related to health, resource management and climate monitoring by leading to better tools for: -Detecting heart disease and cancer -Monitoring and detecting natural resources -Monitoring the environment and climate -Monitoring risk and potential natural disasters. The strong combined research capacity in machine learning for solving real image analysis challenges is a main asset of Visual Intelligence, enabled by the unique interlinking of top research partners and active user partners enabling crucial cross-fertilization between domains. Leveraging top class international cooperation, we will train a large number of researchers and candidates to fill a digital competence void in the Norwegian business and public sector.

Publications from Cristin

Funding scheme:

SFI-Sentre for forskningsdrevet innovasjon

Thematic Areas and Topics

FNs BærekraftsmålFNs BærekraftsmålMål 15 Liv på landFNs BærekraftsmålMål 3 God helseFNs BærekraftsmålMål 14 Liv under vannBransjer og næringerMiljø - NæringsområdeHelseFNs BærekraftsmålMål 7 Ren energi for alleKlimaGlobale klimautfordringerKlimarelevant forskningBransjer og næringerEnergi - NæringsområdeIKT forskningsområdeMenneske, samfunn og teknologiLTP3 HelseLTP3 Klima, miljø og energiBransjer og næringerHelsenæringenBransjer og næringerIKT-næringenLTP3 Klima, polar og miljøIKT forskningsområdeKunstig intelligens, maskinlæring og dataanalysePortefølje Klima og miljøBransjer og næringerMaritim - NæringsområdePortefølje HelseDigitalisering og bruk av IKTOffentlig sektorFornyelse og innovasjon i offentlig sektorKlimaFornyelse og innovasjon i offentlig sektorForskning for fornyelse av offentlig sektorLTP3 Innovasjon i stat og kommuneGrunnforskningAnvendt forskningDelportefølje KvalitetPolitikk- og forvaltningsområderDigitaliseringDelportefølje Et velfungerende forskningssystemIKT forskningsområdeVisualisering og brukergrensesnittInternasjonaliseringLTP3 Fagmiljøer og talenterIKT forskningsområdeDigitalisering og bruk av IKTPrivat sektorLTP3 Høy kvalitet og tilgjengelighetLTP3 Styrket konkurransekraft og innovasjonsevnePolitikk- og forvaltningsområderPortefølje Banebrytende forskningPortefølje Muliggjørende teknologierDelportefølje InternasjonaliseringDigitalisering og bruk av IKTInternasjonaliseringInternasjonalt prosjektsamarbeidLTP3 Et kunnskapsintensivt næringsliv i hele landetBransjer og næringerLTP3 IKT og digital transformasjonPortefølje ForskningssystemetLTP3 Muliggjørende og industrielle teknologierPortefølje Innovasjon