Skip to content
Research Article EISLAB Machine learning group LTU
View original picture , opens in new tab/window

Our research

Published: 1 October 2018

We perform research in various areas of Machine Learning and Artificial Intelligence, including deep learning, pattern recognition, and human computer interaction; with applications in digital humanities, education, document analysis, and Industry 4.0.

Natural Language Processing (NLP)

Our current efforts focus on various machine learning methods for NLP tasks. In a recent workshop we have realized a recognition pipeline for the ambitious task of intent classification on little data where we outperformed

Selected projects where our group or individual members of the group are involve in:

On-going projects

 

AIDIH (Jan. 2020 - Feb. 2023)

The Applied AI DIH North project aims to create a strong innovation system for growth in the AI industry, a Digital Innovation Hub as a base, in collaboration, research, innovation, applied test-driven development, education and clustering. The project lasts for three years and is financed by the EU Regional Development Fund (Tillväxtverket), Luleå University of Technology, Luleå Municipality, Skellefteå Municipality, and Region Norrbotten.

Funding Partner:
- European Regional Development Fund
- Luleå University of Technology
- Luleå Municipality
- Skellefteå Municipality
- Region Norrbotten

Enabling Deep DIA (Jan. 2020 - Dec. 2023)

We  create the missing piece to fundamentally improve automated Document Image Analysis (DIA) systems: A framework for creating databases for DIA; and a massive database of document images for historical DIA. Deep learning methods have disrupted the areas of Computer Vision, Pattern Recognition, and Artificial Intelligence in general. However, despite deep learning being the state of the art in most DIA tasks, their performance is still too low to be useful in practice, especially for domains where not many texts are available. This is mainly due to the absence of large databases for training deep systems. In the first step, we conceptualize and implement a configurable framework for document generation which is adaptable for various domains. The methods of EnDeepDIA will base on logical document layout generation, mathematical models for document deformation, and novel machine learning algorithms for document generation. To demonstrate the feasibility of the EnDeepDIA framework, we  generate a massive public database of historical structured documents (targeting millions of labeled images), including demographic and economic reports. As such EnDeepDIA enhances the capabilities of deep learning for DIA and provides open access services for the Digital Humanities. We will demonstrate the capabilities to break new ground in synthetic database generation for reading systems and related areas of pattern recognition where few training documents are available.

 

Language models for Swedish authorities (Nov. 2019 - Oct. 2022)

This project develops state-of-the-art Swedish language models and applies them to a number of NLP tasks relevant to Swedish authorities. This project provides the tools and prerequisites for Swedish authorities to build and integrate state-of-the-art NLP solutions in their current and future services. The development of state-of-the-art Swedish language models, and techniques for utilizing them, will enable novel types of NLP applications that have the potential to revolutionize the use of NLP in the Swedish public sector.

Funding Partner:
- Vinnova

Nationellt Rymddatalabb (Jun. 2019 - Aug. 2021)

Nationellt Rymddatalabb will be a national knowledge and data hub for Swedish authorities' work on earth observation data and for the development of AI-based analysis of data, generated in space systems. The purpose of the project is to increase the use of data from space for the development of society and industry and for the benefit of the globe.

Our role: 
- Provide advanced image processing methods tailored to the needs of the pilot projects
- Pilot projects include climate adaptation as well as applications of earth observation data in forestry, fishery, and agriculture
- Organise hackathons and user workshops

Partners: 
- Swedish National Space Agency
- AI INNOVATION of Sweden
- RISE
- Luleå University of Technology

 

Nationellt Rymddatalabb 2.0 (Oct. 2021 - Oct. 2023)

The Swedish Space Data Lab was initiated in 2019 as a collaboration project between AI Sweden, the Swedish National Space Agency, RISE Research Institutes of Sweden, and Luleå University of Technology. Space data is used in a wide range of fields. It is indispensable for, among other things, weather forecasts and monitoring the climate, but it is also extremely important for forestry, agriculture, and other fields in which up-to-date information about vegetation and the land surface is needed. The Swedish Space Data Lab is intended to be a national innovation hub for Swedish authorities using Earth observation data, and for the development of AI-based analysis of the data. The purpose of the lab is to enable the increased use of data from space for the development of society for the benefit of the globe. The goal is to get data, technology, and methodology in place to enable the systematic development of space data-based services and applications. The National Space Data Lab 2.0 builds on the experience from the National Space Data Lab. In this follow-up project, we want to expand our Earth Observation work with Artificial Intelligence, annotated data, "edge learning" and business opportunities. We also continue the organization of hackathons and developer events based on the models we develop and train. Moreover, we will work more with the ecosystem around the National Space Data Lab and identify the opportunities for the commercialization of knowledge that exists. 

Funding Partner:
- Vinnova

MetMaskin: Control of metallurgical processes with indirect measurements and machine learning (Nov. 2018 - Oct. 2022)

In this project, we optimize steel manufacturing process steps through the prediction of stirring intensity, which is expected to provide a more efficient steel production that takes place on time and with reduced energy consumption. Prediction of stirring intensity in metallurgical processes is realized by combining the use of measurement technology and machine learning. Better process control is expected to optimize time spent in process steps such as pouring and carbon control in converters.

Funding Partner:
- Vinnova

KnowIT FAST: Knowledge Integration for Fault Severity Estimation​ (Sep. 2019 - Sep. 2022)

Damage to paper machines must be foreseen and can be prevented with artificial intelligence. By integrating methods for sensor data analysis with language technology for analysing written assessments of machine damage, routine chores will be automated, to enable an increased focus on proactive maintenance and to facilitate staff training. This is the goal of the KnowIT FAST project, which is coordinated by the Luleå University of Technology.

Funding Partner: 
- PIIA/Vinnova

Ml-Dev (Jan. 2021 - Dec. 2021)

The work environment 'ML-Dev' of the ML Group enables world-class science and thus has a strong influence on machine learning in society for the welfare of society. This project creates an environment for researchers that enables immediate entry into profound research questions through efficient processes, structures, and knowledge transfer. Due to the working environment's attractiveness, new researchers and worldwide collaborations are brought into the group.

AutoDC: Autonomous datacenters for long term deployment (Oct. 2018 - Oct. 2021)

With growth in the data centre market expected to continue, the cost of operating and maintaining the data centre footprint will increase. The aim of AutoDC is to provide an innovative design framework for autonomous data centres to enable ongoing operation and self-healing independent of contextual interference, e.g. intermittent power failure or overheating, without the need for any human intervention. Due to lower maintenance and operation costs, autonomous data centres can become key enablers of markets in developing countries.

Funding Partner: 
- ITEA3

ChatPal

Current mental health service provision in Northern Sweden cannot meet the rising demand to prevent and manage mental ill-health. There is a lack of digital mental health support for tracking symptoms and for providing treatments and coping strategies at the point of need for 24/7. Traditional one-to-one mental health services supporting people with chronic mental illness as well as mild-to-moderate mental illness is expensive and resource-limited. One-to-one intervention support requires significant travel for clients living in rural areas; hence accessibility to traditional treatments are a particular concern. Given mental ill health remains a
stigma, citizens often feel embarrassed when setting up appointments with a support person. Mental ill-health, however, particularly when left untreated can decrease people's functional capacity and create obstacles for participation. Therefore our objective is to use already existing, and refined methods in developing new conversational agents (both text- and speech-based) users can communicate to without the risk of being stigmatized and that can provide support as well rudimentary assessment of emotional and mental well-being.

 

Finished

HisDoc III

In HisDoc III we target historical document classification for large amounts of uncategorized facsimiles with the intent to provide new capabilities for researchers in the Digital Humanities. In particular, we will address the task of categorizing document images with respect to content, language, script, and layout. To do so, we will leverage the expertise gained from our previous projects HisDoc and HisDoc 2.01. In HisDoc we have shown that historical Document Image Analysis (Dia) can be effectively applied to extract layout structures and textual transcriptions and in the current HisDoc 2.0 project we successfully retrieved additional paleographic information. The novel contributions of HisDoc III will be complemented by these methods to cope with large document collections.

iMuSciCA

iMuSciCA is a pioneering approach using music for fostering creativity and deeper learning. The DIVA group introduces pen- and gesture-based interaction for music co-creation and sound analytics.

LTU-ServiceDesk-Chatbot

A collaborative project with the LTU Service Desk with the aim to build a chatbot to facilitate the provision of solutions for students/staff of the university based on their requests currently handled by the university ServiceDesk. The chatbot will automatically handle the requests and redirect LTU users to possible solutions through textual conversations by using a web interface on the browser.

Partners: 
- LTU Service Desk

SUN FLA: Fusing Machine Learning and Computer Vision Techniques for Automatic Drill Core Analysis on Visual and Compositional Data - ML4DrillCore (Feb. 2021 - Nov. 2021)

Quantitative and qualitative geological models of mineral deposits form the basis of the mining value chain, including all subsequent decisions on valuation, mining method, processing method, and measures to alleviate the environmental impact of mining. The main input data in such models are derived from the analysis of drill cores. Owing to the complexity of geological materials coupled with time constraints, only a fraction of the information stored in the rocks is collected. Subjectiveness in logging of textural features furthermore leads to that results may differ between observers, inducing additional complexity. These uncertainties and confusion impact negatively on the models’ reliability, which may, in turn, have adverse effects further down the value chain. ML4DrillCore aims at improving the analysis and evaluation of drill core data with the combined use of deep learning techniques, like for example computer vision techniques on visual data and standard machine learning methods on compositional data.

Funding Partner: 
- SUN seed projects