Context and dependencies

Background

The strength of machine learning methods is the ability to learn from data rather than using predefined models. For complex data there is however a need to integrate the best of these two worlds to enable integration of physical or geometrical models, dependencies, and prior knowledge, as well as the exploitation of multiple complex image modalities simultaneously.

Challenges

Current deep learning systems for image analysis depend on individual pixel information, capturing dependencies solely via the convolution neighborhood.

This means that the ability to incorporate context and prior knowledge, e.g. about topology or boundaries, is limited. The ability to conform to physical models, and principles governing the image data generation and its properties is also limited, including modelling of temporal dependencies and processes. In order to make deep learning based computer vision systems ubiquitous and applicable also for complex, sparsely labelled image data, there is a need for visual intelligence that can easily be adapted to new, non-standard data sources with few labelled training samples.

Main objective

To develop new methodology to exploiting context, dependencies and prior knowledge in deep learning.

Related news

Paper published in International Journal of Remote Sensing
June 28, 2021

Qinghui Liu, Michael Kampffmeyer, Robert Jenssen and Arnt-Børre Salberg have published their paper Self-constructing graph neural networks to model long-range pixel dependencies for semantic segmentation of remote sensing images in International Journal of Remote Sensing.

New paper published in Machine Learning (2021)
June 24, 2021

Ahcène Boubekki, Michael Kampffmeyer, Ulf Brefeld and Robert Jenssen have published their paper Joint optimization of an autoencoder for clustering and embedding in journal Machine Learning (2021).

Stream our latest seminars
May 27, 2021

Did you miss any of our recent seminars? When we host seminars and events we often record relevant talks and presentations and make them available at our youtube channel. You can access all our content through our "outreach" page.

Annual report 2020
May 20, 2021

SFI Visual Intelligence has published the annual report for 2020. The report is approved by the Visual Intelligence board and available for download as a pdf document under "publicity".

Visual Intelligence Graduate School (VIGS)
May 20, 2021

SFI Visual Intelligence is organizing a graduate school for early career research fellows connected to Visual Intelligence. VIGS aims at connecting research fellows across our different research institutions to build social and professional networks.

Paper accepted at CVPR 2021
March 19, 2021

We are proud to announce that the paper “Reconsidering Representation Alignment for Multi-view Clustering” by Daniel J. Trosten, Sigurd Løkse, Robert Jenssen and Michael Kampffmeyer was recently accepted at the Conference on Computer Vision and Pattern Recognition 2021!

Official opening of Visual Intelligence research centre!
January 19, 2021

January 14, 2021 the official opening of SFI Visual Intelligence will be organized at the UiT - The Arctic University of Norway. Anne Husebekk, the rector of UiT will be giving a speech at the opening ceremony.

Northern Lights Deep Learning Workshop 2021
January 19, 2021

NLDL 2021 will be a digital conference hosted by the UiT Machine Learning Group and Visual Intelligence January 18-20. The program includes a Mini Deep Learning School the 18th and is followed by a tight program the rest of the week.

A new Centre for Research-based Innovation
January 19, 2021

Visual Intelligence will be one of the new SFIs funded by the Research Council of Norway. The center will run over a period of eight years and will form a collaboration between businesses and research institutions in Norway.

Visual Intelligence is officially opened!
January 19, 2021

The official opening of SFI Visual Intelligence was successfully arranged as a digital event today. We are now ready to commence our research and innovation to tackle some of the large challenges in deep learning and AI, along with our partners.

Related projects

Opening the black box of AI
January 19, 2021
Deep learning and AI models must become interpretable, explainable and reliable before they can be utilized in complex domains.
Modelling continuity in seismic data
December 17, 2020
Visual intelligence is collaborating with Equinor to develop models that can exploit seismicdata and model the continuity of the subsurface.
New methods for automatic change detection in aerial images
December 17, 2020
A collaboration with Terratec to develop deep learning methods to automatically detect changes when updating an existing map database.