Welcome to the Swedish symposia on deep learning and image analysis!

SSDL/SSBA 2022 goes virtual!

Due to the pandemic, the symposium can not be held on site in Uppsala, but will be a virtual event. Participation will be free-of-charge for all. SSDL/SSBA 2022 thereby marks the continuation of a 39-year-old tradition of gathering researchers in the field of image analysis and pattern recognition who are active in Sweden to share their latest findings and socialize in an annual national forum.

SSDL is an important Swedish forum for leading research groups in industry and academia to meet and discuss the latest trends and developments in deep learning and related areas. SSDL 2022 will feature invited talks by leading researchers in deep learning as well as oral and poster presentations of submitted papers and abstracts.

The SSBA symposium is the premier Swedish event where researchers, industrial professionals and students gather to learn about the recent developments in the areas of image processing, computer vision, pattern recognition and related fields. SSBA 2022 features keynote speakers and oral presentations and posters of submitted papers.

SSDL/SSBA 2022 is organized by the Department of Information Technology at Uppsala University, Centre for Image Analysis, and Swedish Society for Automated Image Analysis.

Important dates

SSBA paper submission deadline Passed
SSDL abstract/paper submission deadline Passed

Registration deadline


Conference dates
Deep Learning Symposium (SSDL) 14 March 12.00-18:00
Image Analysis Symposium (SSBA) 15 - 16 March


Keynote speakers

Amy Loutfi
  • Amy Loutfi
  • Örebro University

  • Deep Learning in Collaborative Human Robot Interaction Scenarios.
  • I will speak about how the need for robots and other artificial agents to truly understand human instruction increases. This understanding involves the ability for robots and artificial agents to gain semantic meaning that refer to physical objects, actions, and concepts, and connect this to the perceptual information coming from the sensors embedded on the agent. This challenge in part deals with the symbol grounding problem, that is, how to ground symbols into something other than symbols. Further with the advent of advances within representation learning, due to many of the advances within sub-symbolic architectures there has been an opportunity to examine how the grounding can be learned using large (and simulated) datasets. Further, this talk will describe how the semantic representations can also further guide sub-symbolic learning. This talk will highlight the research at Örebro University where semantics are learned, integrated and used for collaborative robot systems that work both for and with humans.

Orcun Göksel
  • Orcun Göksel
  • Uppsala University

  • Deep Learning in Medical Ultrasound Imaging: Simulation and Image Reconstruction
  • Ultrasound is a cost-effective, portable, and real-time imaging modality, involving minimal safety risks for the patient and the sonographer. In this talk, our recent advances in deep learning for ultrasound image analysis will be presented in two different settings. In the first setting, learning image translation from unpaired samples will be demonstrated for the computational simulation of ultrasound imaging, which is essential for training sonographers for complex imaging tasks in a simulated virtual-reality environment. In a second setting, our solutions with deep variational networks in image reconstruction will be demonstrated, in particular for loop-unrolling iterative solutions of inverse problems that occur in ultrasound computed tomography. This is presented for a novel imaging approach for tissue speed-of-sound, which is a promising new modality for breast cancer and muscular degeneration.

    Orcun Goksel is associate professor at the Department of Information Technology at Uppsala University. He leads the research group Computer-assisted Applications in Medicine, as part of Uppsala Medtech Science and Innovation Centre and the Centre for Image Analysis. He received bachelor degrees in computer science and electrical engineering. Following a PhD degree from University of British Columbia, in Vancouver, Canada, he held a Swiss National Science Foundation Professorship at ETH Zurich, Switzerland, before joining Uppsala University. His research interests include machine learning, computer vision, ultrasound imaging in biomedical applications, biomechanical tissue characterization, image reconstruction, image-guided therapy, and patient-specific modelling and simulation. By devising novel imaging and image analysis techniques and developing them for clinical translation, his research efforts push the boundaries of diagnostic and surgical procedures as well as minimally-invasive interventions.

Cynthia Rudin
  • Cynthia Rudin
  • Duke University

  • Interpretable Neural Networks for Computer Vision: Clinical Decisions that are Computer-Aided, not Automated
  • Let us consider a difficult computer vision challenge. Would you want an algorithm to determine whether you should get a biopsy, based on an x-ray? That's usually a decision made by a radiologist, based on years of training. We know that algorithms haven't worked perfectly for a multitude of other computer vision applications, and biopsy decisions are harder than just about any other application of computer vision that we typically consider. The interesting question is whether it is possible that an algorithm could be a true partner to a physician, rather than making the decision on its own. To do this, at the very least, we would need an interpretable neural network that is as accurate as its black box counterparts. In this talk, I will discuss two approaches to interpretable neural networks: (1) case-based reasoning, where parts of images are compared to other parts of prototypical images for each class, and (2) neural disentanglement, using a technique called concept whitening. The case-based reasoning technique is strictly better than saliency maps, and the concept whitening technique provides a strict advantage over the posthoc use of concept vectors. Here are the papers I will discuss:

    This Looks Like That: Deep Learning for Interpretable Image Recognition. NeurIPS spotlight, 2019. https://arxiv.org/abs/1806.10574

    IAIA-BL: A Case-based Interpretable Deep Learning Model for Classification of Mass Lesions in Digital Mammography, 2021. https://arxiv.org/abs/2103.12308

    Concept Whitening for Interpretable Image Recognition. Nature Machine Intelligence, 2020. https://rdcu.be/cbOKj

    Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and use Interpretable Models Instead, Nature Machine Intelligence, 2019. https://rdcu.be/bBCPd

    Interpretable Machine Learning: Fundamental Principles and 10 Grand Challenges, 2021 https://arxiv.org/abs/2103.11251

David J.T. Sumpter
  • David J.T. Sumpter
  • Uppsala University

  • Why mathematical modelling needs to be more creative and unconventional
  • One of the challenges for modellers trying to understand complex biological and social systems is that cultures build up around what are ‘interesting’ research topics. My own academic career (starting 1996) has seen crazes for self-organisation, evolutionary game theory, network science and now machine learning. All of these approaches brought interesting theoretical and applied results, but seen over the longer term, we also notice a recurring problem: Craze X takes on a greater importance than the Science itself. Applications are all re-envisioned in light of Craze X and theoretical debates become about the (often meaningless) details of Craze X. With reference to my own research and popular science writing, I describe how I think we can escape this trap, through a combination of two things: extremely focused applications and extremely unfocussed theoretical investigations. I illustrate the former with my work on modelling football. I illustrate the latter with an example from Artificial Life. I will talk about how wrongly focusing on the middle ground – the development of computational and mathematical theory – can lead to dangerous and unnecessary hype, which can then have negative social impacts.

Marcus Liwicki
  • Marcus Liwicki
  • Luleå University of Technology

  • Historical Document Analysis – history, trends, and challenges
  • I will speak about the development in the field of historical document image analysis during the last two decades (before and after deep learning), the current development, as well as open challenges. Historical document image analysis remains one of the most challenging topics for the document image analysis research community. Historical documents differ from the ordinary documents due to the presence of different artifacts. Issues such as poor conditions of the documents, texture, noise and degradation, large variability of page layout, page skew, random alignment, variety of fonts, presence of embellishments, variations in spacing between characters, words, lines, paragraphs and margins, overlapping object boundaries, superimposition of information layers, etc bring complexity issues in analyzing them. Most methods currently rely on deep learning based methods, including Convolutional Neural Networks and Long Short-Term Memory Networks. As those methods are data-hungry, a recent trend is to generate more training data with various methods. In our VR project, we are dealing with this challenge and working on a systematic approach towards training sets with millions of documents.

Virginie Uhlmann
  • Virginie Uhlmann
  • European Bioinformatics Institute (EMBL-EBI) in Cambridge, UK

  • Deep learning for bioimage analysis: how far we got, and where we are heading
  • Deep learning has transformed the way large and complex microscopy image datasets can be processed, reshaping what is possible in bioimage analysis. In this talk, we will characterize the unique features and challenges of microscopy data, and review how deep learning methods have successfully been adapted from computer vision and medical imaging to efficiently solve recurring bioimage analysis problems. We will also discuss the future of deep learning applied to biology, and explore how state-of-the-art approaches in machine learning have the potential to transform our understanding of biological systems through new analysis and modelling strategies that integrate multimodal inputs in space and time.

Submissions, SSDL (Password protected)

The submissions for the SSDL poster session is listed below. Some of the submissions are also presented as regular talks.

Submissions, SSBA (Password protected)

The submissions for the SSBA poster session is listed below. Some of the submissions are also presented as regular talks.

All SSBA submissions are also available in the SSBA proceedings.


Venue: Online

If you have any questions please contact us on ssbassdl2022@ssba.org.se


Industry support

Last Updated Date: 14 March 2022