Research
Helga Ingimundardóttir
Research
Machine learning and optimisation: learning good decision rules from data instead of hand-crafting them.
Grants
Project grants
-
2024–26
HiDef Textiles: Optimising Textile Processes with AI
University of Iceland Research Fund
Project grant merging AI, industrial engineering, computer science and textile craftsmanship.
The goal is to democratise technology and programming through textiles, integrating arts and sciences. We are developing a sustainable learning platform to share industrial best practice, with an emphasis on women in STEM.
-
2025
HiDef Textiles: Technology-Enabled Knitting Machines
University of Iceland: Grants to Support Public Outreach
ISK 1,430,000 grant to share research and specialist knowledge with the public.
The grant supports an interactive website that shares the research software and the smart knitting machine’s development with the public. The project connects technology, sustainability and Icelandic textile heritage, and invites people to explore and try Icelandic patterns.
-
2024
Icelandic Centre for Research, Student Innovation Fund
Funding to upgrade a 1990s knitting machine to modern autonomous standards — three BSc graduates for three months each.
Part of HiDef Textiles. The project combines sustainable technology, textile innovation, STEAM education, AI and IoT. By the end of the summer a functional knitting machine was ready to present at local conferences.
-
2012
Grant for Teaching Development
University of Iceland, Academic Affairs Fund
The project Breytt dæmatímafyrirkomulag hjá VON — a new tutorial format at SENS.
A grant to improve undergraduate homework assignments in Engineering and Natural Sciences: a simpler method that reduces teacher workload while giving students better feedback. The work was done as part of the course Reflective practice and professional development.
-
2009–12
Postgraduate Scholarship
University of Iceland Research Fund
Three-year stipend for doctoral studies.
Nothing to show at this detail level.
Mobility grants
-
2024
Erasmus+ Higher Education Mobility Grant
ERASMUS+
Study visit to Aalborg University and DTU on project-based learning and curriculum integration.
The mobility enhances the University of Iceland teaching methodology by learning from Aalborg University project-based learning and DTU curriculum integration. The collaboration improves the curriculum, smooths BSc to MSc transitions, and strengthens international partnerships.
-
2010
Higher Education Mobility Grant
Fundusz Stypendialny i Szkoleniowy (FSS)
Mobility grant to visit Politechnika Śląska in Gliwice, Poland.
-
2009
Postgraduate Scholarship
French Embassy, Reykjavík
Awarded to Icelandic students pursuing a Masters degree in France.
Nothing to show at this detail level.
Publications
Journal articles
-
2025
Viðskiptagreind sem brú milli náms og starfs
ArticleTímarit Kennslumiðstöðvar Háskóla Íslands, 12(1), pp. 46–49, 2025
English summary (translated from the Icelandic article): The article describes the redesign of a Business Intelligence course to connect university learning with workplace practice. Drawing on student evaluations, alumni feedback, and conversations with industry partners, it explains how team-based learning, flipped classrooms, GitHub, and authentic business data were used to make project work more coherent and professionally relevant. It reflects on adapting teaching methods to the instructor's experience outside academia and on balancing student learning with useful outcomes for partner organisations.
-
2021Article
Nature Genetics, 53(6), pp. 779 – 786, Springer Nature, 2021
Long-read sequencing (LRS) promises to improve the characterization of structural variants (SVs). We generated LRS data from 3,622 Icelanders and identified a median of 22,636 SVs per individual (a median of 13,353 insertions and 9,474 deletions). We discovered a set of 133,886 reliably genotyped SV alleles and imputed them into 166,281 individuals to explore their effects on diseases and other traits. We discovered an association of a rare deletion in PCSK9 with lower low-density lipoprotein (LDL) cholesterol levels, compared to the population average. We also discovered an association of a multiallelic SV in ACAN with height; we found 11 alleles that differed in the number of a 57-bp-motif repeat and observed a linear relationship between the number of repeats carried and height. These results show that SVs can be accurately characterized at the population scale using LRS data in a genome-wide non-targeted approach and demonstrate how SVs impact phenotypes. © 2021, The Author(s), under exclusive licence to Springer Nature America, Inc.
-
2021
Ratatosk: hybrid error correction of long reads enables accurate variant calling and assembly
Article Open accessGenome Biology, 22(1), Springer Nature, 2021
A major challenge to long read sequencing data is their high error rate of up to 15%. We present Ratatosk, a method to correct long reads with short read data. We demonstrate on 5 human genome trios that Ratatosk reduces the error rate of long reads 6-fold on average with a median error rate as low as 0.22 %. SNP calls in Ratatosk corrected reads are nearly 99 % accurate and indel calls accuracy is increased by up to 37 %. An assembly of Ratatosk corrected reads from an Ashkenazi individual yields a contig N50 of 45 Mbp and less misassemblies than a PacBio HiFi reads assembly. © 2021, The Author(s).
-
2018Article
Journal of Scheduling, 21(4), pp. 413 – 428, Springer Nature, 2018
Dispatching rules can be automatically generated from scheduling data. This paper will demonstrate that the key to learning an effective dispatching rule is through the careful construction of the training data, xi(k),yi(k)k=1K∈D, where (i) features of partially constructed schedules xi should necessarily reflect the induced data distribution D for when the rule is applied. This is achieved by updating the learned model in an active imitation learning fashion; (ii) yi is labelled optimally using a MIP solver; and (iii) data need to be balanced, as the set is unbalanced with respect to the dispatching step k. Using the guidelines set by our framework the design of custom dispatching rules, for a particular scheduling application, will become more effective. In the study presented three different distributions of the job-shop will be considered. The machine learning approach considered is based on preference learning, i.e. which dispatch (post-decision state) is preferable to another. © 2017, Springer Science+Business Media New York.
-
2018
Insights into imprinting from parent-of-origin phased methylomes and transcriptomes
ArticleNature Genetics, 50(11), pp. 1542 – 1552, Nature Publishing Group, 2018
Imprinting is the preferential expression of one parental allele over the other. It is controlled primarily through differential methylation of cytosine at CpG dinucleotides. Here we combine 285 methylomes and 11,617 transcriptomes from peripheral blood samples with parent-of-origin phased haplotypes, to produce a new map of imprinted methylation and gene expression patterns across the human genome. We demonstrate how imprinted methylation is a continuous rather than a binary characteristic. We describe at high resolution the parent-of-origin methylation pattern at the 15q11.2 Prader–Willi/Angelman syndrome locus, with nearly confluent stochastic paternal methylation punctuated by ‘spikes’ of maternal methylation. We find examples of polymorphic imprinted methylation unrelated (at VTRNA2-1 and PARD6G) or related (at CHRNE) to nearby SNP genotypes. We observe RNA isoform-specific imprinted expression patterns suggestive of a methylation-sensitive transcriptional elongation block. Finally, we gain new insights into parent-of-origin-specific effects on phenotypes at the DLK1/MEG3 and GNAS loci. © 2018, The Author(s), under exclusive licence to Springer Nature America, Inc.
-
2011
Detection of fouling in a cross-flow heat exchanger using wavelets
ArticleHeat Transfer Engineering, 32(3-4), pp. 349 – 357, Taylor and Francis Ltd., 2011
Detection of fouling in a heat exchanger experiencing perfect steady-state conditions is not very difficult. But the challenge is to detect fouling when all inputs (inlet temperature of the fluids and the mass flow rates) are simultaneously varying. In this paper it has been considered that the mass flow rates can vary in a ratio of 2, and that the inlet temperatures can vary by about ±20%. This first approach is dedicated to show the feasibility of using the wavelet transform. It has been considered that getting simulated data is the best way. In fact, it is then possible to introduce an arbitrary fouling factor. Thus, in the first part of the paper the model of the heat exchanger is presented. It is developed using Simulink. The validation is carried out on an electrical heater, for which it is possible to find an analytical solution for transient states. It is also shown that steady states are accurately computed over a large range of the number of transfer units and heat capacity rate ratios. Then a brief overview of the wavelet transform is given. Then basic examples show that the wavelet transform can help to find the trend of time series. It is then applied to the analysis of the "wavelet- transformed" effectiveness of the heat exchanger. This analysis is carried out on a sliding observation window (to be able to detect fouling on-line). It is shown that fouling is detected at a very early stage. Copyright © Taylor and Francis Group, LLC.
Nothing to show at this detail level.
Peer-reviewed conference proceedings
-
2026
A Scalable Matheuristic for Routing Capacity-Constrained Groundfish Surveys
Conference paperLearning and Intelligent Optimization. LION 2026, Springer International Publishing, 2026, Accepted
Bottom trawl groundfish surveys are planned around a fixed set of sampling stations, but the order of visits and the timing of port returns strongly affect total travel distance and operational feasibility. Although mixed-integer programming (MIP) models can encode capacity and port-call constraints directly, solving the full model to optimality is often impractical at realistic scales. We propose a matheuristic that combines fast tour construction with repeated calls to a time-limited capacity MIP subproblem. Starting from an initial segmented tour, the algorithm iteratively examines boundaries between adjacent segments separated by a port visit and solves a restricted two-segment capacity MIP on the union of stations in the two segments, with fixed endpoints and a single intermediate port call. The MIP reallocates stations between the two segments subject to capacity, and the best incumbent found within a time limit is accepted if it improves total distance. We evaluate the approach under four initialization strategies and a range of MIP time limits. Across scenarios, the method consistently improves baseline plans and provides a fast, reproducible tool for strategic planning and scenario comparison. These results suggest that restricted-MIP local improvement is an effective mechanism for producing high-quality survey routes under realistic operational constraints.
-
2026
Designing Authentic Industry-Engaged Assessment for Professional Competence in Business Intelligence
Conference paper Open accessProceedings of the 22nd International CDIO Conference, pp. 211 – 220, 2026
Engineering programs increasingly seek to align student learning with professional practice, yet meaningful industry collaboration often exposes misalignment between academic assessment norms and workplace expectations. This paper presents a CDIO Implementation case describing the redesign of an undergraduate business intelligence course to strengthen industrial relevance while supporting student professional development. The redesign integrated semester-long, industry-inspired project work structured around team-based learning and competence-oriented assessment, emphasizing authentic data, reproducible workflows, and professional collaboration tools. A single industrial partner contributed real-world datasets, problem framing, and feedback, enabling students to engage with constraints and quality standards characteristic of professional practice. Assessment and feedback were redesigned to balance team performance with individual contribution through visible artifacts (e.g. version control activity) and structured reflection, addressing persistent challenges related to workload, accountability, and fairness. Evidence from iterative course evaluations, student reflections, graduate feedback, and industry partner input indicates increased student engagement, improved coherence across learning activities, and clearer alignment between assessment criteria and professional expectations. At the same time, the collaboration revealed tensions between academic tolerance for developmental errors and industry demands for reliability and trustworthiness, highlighting the importance of explicitly preparing students to communicate differently with academic and professional audiences. The paper reflects on these tensions and outlines design principles for industry-engaged CDIO courses, including expectation alignment early in the course, sustained use of a shared industrial context, and assessment practices that foreground professional standards without sacrificing formative learning space. The case offers transferable insights for CDIO programs seeking to leverage industrial collaboration as a catalyst for student professional development through innovative assessment and feedback. © 2026 22nd International CDIO Conference. All right reserved.
-
2026
Using Pull Requests to Make Collaboration Visible in CDIO Project-Based Courses
Conference paper Open accessProceedings of the 22nd International CDIO Conference, pp. 296 – 305, 2026
Assessing collaboration, individual accountability, responsiveness to feedback, and reproducibility remains a persistent challenge in open-ended, team-based project courses. Final deliverables often mask how work evolved, how responsibilities were shared, and how decisions were negotiated. This paper presents an assessment design in which pull requests (PRs), a mechanism for reviewing and integrating changes to shared project work, serve as the central mechanism for making collaborative practices visible and assessable within CDIO-aligned project-based courses. Rather than treating GitHub as a programming topic or tooling exercise, the approach positions PRs as assessment infrastructure that binds technical change to process evidence through review discussions, explicit responses to comments, and documented handover. The implementation spans two intentionally sequenced courses in Industrial Engineering. A second-year course introduces structured collaboration practices through short, skills-focused project cycles. A subsequent advanced course assumes GitHub fluency and organizes all project work within a single persistent repository maintained across the semester. Within this environment, PRs function as the primary unit of feedback, peer review, and evaluation. Students make their reasoning, revisions, and communication visible through routine workflow activity, and unresolved feedback carries forward across project cycles into capstone evaluation, enabling cumulative rather than fragmented assessment. Experience across multiple offerings shows that PR-centered assessment improves the scope and specificity of peer review, strengthens documentation and reproducibility practices, and fosters shared responsibility for evolving team artifacts. Students initially expressed concern about perceived overhead, but review quality, documentation habits, and disciplined handover improved markedly as the workflow became routine. Graduating students later identified PR-based communication and accountability as directly transferable to professional settings. The paper highlights key design decisions, trade-offs, and sequencing choices, offering practical and transferable guidance for educators seeking assessment methods that make collaboration, responsiveness, and reproducibility genuinely visible in CDIO project-based courses. © 2026 22nd International CDIO Conference. All right reserved.
-
2024
Evolving Submodels for Column Generation in Cutting and Packing for Glulam Production
Conference paper2024 IEEE Congress on Evolutionary Computation (CEC), IEEE, 2024
This study introduces a novel optimization approach for the pressing phase in glued laminated timber (glulam) production, integrating column generation techniques with an evolution-ary algorithm within a real-world glulam factory context. At its core, an evolutionary strategy refines parameters in a sub model dedicated to developing cutting patterns via column generation. These patterns are then utilized in a separate, overarching 2D packing problem, targeting the optimization of the pressing process while adhering to pressing capacity constraints. The solutions' effectiveness feeds back into the evolutionary strategy, guiding the optimization of its parameters. This iterative, multi-layered framework presents a significant advancement in glulam manufacturing, aiming to enhance production efficiency, reduce waste, and expedite customer order fulfillment. © 2024 IEEE.
-
2016
Evolutionary learning of linear composite dispatching rules for scheduling
Conference paperStudies in Computational Intelligence, 620, pp. 49 – 62, Springer Verlag, 2016
A prevalent approach to solving job shop scheduling problems is to combine several relatively simple dispatching rules such that they may benefit each other for a given problem space. Generally, this is done in an ad-hoc fashion, requiring expert knowledge from heuristics designers, or extensive exploration of suitable combinations of heuristics. The approach here is to automate that selection by translating dispatching rules into measurable features and optimising what their contribution should be via evolutionary search. The framework is straight forward and easy to implement and shows promising results. Various data distributions are investigated for both job shop and flow shop problems, as is scalability for higher dimensions. Moreover, the study shows that the choice of objective function for evolutionary search is worth investigating. Since the optimisation is based on minimising the expected mean of the fitness function over a large set of problem instances which can vary within the set, then normalising the objective function can stabilise the optimisation process away from local minima. © Springer International Publishing Switzerland 2016.
-
2015
Generating training data for learning linear composite dispatching rules for scheduling
Conference paperLecture Notes in Computer Science, 8994, pp. 236 – 248, Springer International Publishing, 2015
A supervised learning approach to generating composite linear priority dispatching rules for scheduling is studied. In particular we investigate a number of strategies for how to generate training data for learning a linear dispatching rule using preference learning. The results show, that when generating a training data set from only optimal solutions, it is not as effective as when suboptimal solutions are added to the set. Furthermore, different strategies for creating preference pairs is investigated as well as suboptimal solution trajectories. The different strategies are investigated on 2000 randomly generated problem instances using two different problem generator settings. © Springer International Publishing Switzerland 2015.
-
2014
Evolutionary Learning of Weighted Linear Composite Dispatching Rules for Scheduling
Conference paperInternational Joint Conference on Computational Intelligence, 1, pp. 59 – 67, Science and Technology Publications, Lda, 2014
A prevalent approach to solving job shop scheduling problems is to combine several relatively simple dispatching rules such that they may benefit each other for a given problem space. Generally, this is done in an ad-hoc fashion, requiring expert knowledge from heuristics designers, or extensive exploration of suitable combinations of heuristics. The approach here is to automate that selection by translating dispatching rules into measurable features and optimising what their contribution should be via evolutionary search. The framework is straight forward and easy to implement and shows promising results. Various data distributions are investigated for both job shop and flow shop problems, as is scalability for higher dimensions. Moreover, the study shows that the choice of objective function for evolutionary search is worth investigating. Since the optimisation is based on minimising the expected mean of the fitness function over a large set of problem instances which can vary within the set, then normalising the objective function can stabilise the optimisation process away from local minima. © 2014 SCITEPRESS (Science and Technology Publications, Lda.).
-
2012Conference paper
Lecture Notes in Computer Science, 7219 LNCS, pp. 408 – 412, 2012
Many heuristic methods have been proposed for the job-shop scheduling problem. Different solution methodologies outperform other depending on the particular problem instance under consideration. Therefore, one is interested in knowing how the instances differ in structure and determine when a particular heuristic solution is likely to fail and explore in further detail the causes. In order to achieve this, we seek to characterise features for different difficulties. Preliminary experiments show there are different significant features that distinguish between easy and hard JSSP problem, and that they vary throughout the scheduling process. The insight attained by investigating the relationship between problem structure and heuristic performance can undoubtedly lead to better heuristic design that is tailored to the data distribution under consideration. © 2012 Springer-Verlag.
-
2011
Sampling strategies in ordinal regression for surrogate assisted evolutionary optimization
Conference paperInternational Conference on Intelligent Systems Design and Applications, ISDA, pp. 1158 – 1163, 2011
In evolutionary optimization surrogate models are commonly used when the evaluation of a fitness function is computationally expensive. Here the fitness of individuals are indirectly estimated by modeling their rank with respect to the current population by use of ordinal regression. This paper focuses on how to validate the goodness of fit for surrogate models during search and introduces a novel validation/updating policy for surrogate models, and is illustrated on classical numerical optimization functions for evolutionary computation. The study shows that for validation accuracy it is sufficient for the approximate ranking and true ranking of the training set to be sufficiently concordant or that only the potential parent individuals should be ranked consistently. Moreover, the new validation approach reduces the number of fitness evaluation needed, without a loss in performance. © 2011 IEEE.
-
2011
Supervised learning linear priority dispatch rules for job-shop scheduling
Conference paperLecture Notes in Computer Science, 6683 LNCS, pp. 263 – 277, 2011
This paper introduces a framework in which dispatching rules for job-shop scheduling problems are discovered by analysing the characteristics of optimal solutions. Training data is created via randomly generated job-shop problem instances and their corresponding optimal solution. Linear classification is applied in order to identify good choices from worse ones, at each dispatching time step, in a supervised learning fashion. The method is purely data-driven, thus less problem specific insights are needed from the human heuristic algorithm designer. Experimental studies show that the learned linear priority dispatching rules outperforms common single priority dispatching rules, with respect to minimum makespan. © Springer-Verlag Berlin Heidelberg 2011.
Nothing to show at this detail level.
Peer-reviewed extended abstracts
-
2026
Endurhönnun Upplýsingaverkfræði: Stigvaxandi þróun faglegra vinnubragða
Ráðstefna Kennsluakademíu opinberu háskólanna, Veröld Háskóla Íslands, Reykjavík, 2026, Submitted
This project describes the ongoing redesign of Information Engineering (IDN302G), a compulsory second-year course in the undergraduate industrial engineering programme. The course develops computer and data literacy, with an emphasis on data acquisition, data processing, GitHub, and communicating results. Course materials and teaching arrangements are published on the open Information Engineering course website.
The redesign draws on lessons from the more advanced elective Business Intelligence course (IDN610M), taken by third-year undergraduate and master's students. In that course, students work on more open-ended analytical projects, use more complex real-world data from an industry partner, apply practical AI methods, and communicate findings to different stakeholders, including company managers. The aim is to teach more of the foundations of professional workflows earlier and in a beginner-friendly way, leaving more room in the advanced course to tackle complex datasets, open-ended problems, and professional communication.
Information Engineering therefore focuses on the workflow itself: formulating questions, acquiring and organising data, analysing it, making the work reproducible, and communicating results clearly. The course is also designed so that these practices transfer to many other courses in the programme, not only Business Intelligence. Students work with open, accessible datasets rather than directly with a company or institution. This creates an accessible, professionally oriented introduction to data-driven engineering practice.
From a pedagogical perspective, this design can be understood as a form of scaffolding, in which support is adapted to students' level, gradually withdrawn, and responsibility for the workflow transferred to them (van de Pol et al., 2010). The alignment of learning outcomes, recurring assignments, and assessment criteria also draws on constructive alignment: student activities and assessment should elicit the performance described by the learning outcomes (Biggs, 1996).
-
2026
Erindrekinn vinnur, nemandinn ber ábyrgð
Ráðstefna Kennsluakademíu opinberu háskólanna, Veröld Háskóla Íslands, Reykjavík, 2026, Submitted
AI agents are changing how artificial intelligence is used in software development, data analysis, and other forms of knowledge work. Unlike conventional chatbots or automatic code completion, agents can be given a goal, inspect a project's context, use tools, and carry out multi-step actions such as editing files, running tests, and preparing a pull request. Surveys suggest that use of such systems is growing rapidly, although their adoption remains at an experimental stage in many organizations. In a global McKinsey survey, 62% of respondents said their organizations were at least experimenting with AI agents, while few had implemented them at scale (McKinsey & Company, 2025). In a JetBrains survey of software professionals, 74% said they used specialized AI tools in development, including coding assistants and agents (JetBrains Research, 2026).
This development calls for new professional skills. It is no longer enough to know how to prompt AI to produce code or analysis. Users must also be able to scope a task, direct the workflow, assess and verify results, and take responsibility for the final product.
In autumn 2026, the use of agents was deliberately integrated into IDN302G Information Engineering, a compulsory second-year undergraduate course in industrial engineering at the University of Iceland. The course develops computer and data literacy with an emphasis on professional workflows: formulating questions, acquiring and processing data, managing work in GitHub, making analyses reproducible, and communicating results clearly (Ingimundardóttir, 2026).
The use of agents is part of a broader redesign of the course. The aim is not merely to teach students how to use new AI tools, but to help them apply these tools within a professional workflow: scope tasks, delegate appropriate work to an agent, review its output, and take responsibility for the results. The challenge is to benefit from the execution capabilities agents provide without allowing automation to displace the thinking, decision-making, and oversight that the course aims to develop.
-
2026
Sýnileg samvinna með tólum atvinnulífsins
11. kennsluráðstefna Kennslu- og upplýsingatæknimiðstöðvar Háskólans á Akureyri (KHA), Zoom, 2026
Á tímum þar sem háskólar leita leiða til að efla gagnsæi, ábyrgð og fagmennsku í námsmati geta vinnulag og tól úr atvinnulífinu skapað ný, áhrifarík tækifæri í kennslu. Fjallað verður um hvernig innleiðing pull request (PR) vinnulags, GitHub Classroom og samfellds, endurtekins vinnuferils hefur breytt verkefnadrifinni kennslu í tveimur námskeiðum í upplýsingaverkfræði og viðskiptagreind við Háskóla Íslands. PR miðað mat gerir samvinnu nemenda sýnilega á öllum stigum: rökstuðning, jafningjarýni, viðbrögð við endurgjöf, ábyrgð á gögnum og skipulagða afhendingu. Með því að festa endurgjöf við PR þræði flytjast athugasemdir og umbótakröfur milli lota, sem skapar samfellu, dýpri ígrundun og réttlátara námsmat. Nemendur æfa þannig vinnubrögð sem tíðkast í atvinnulífi, þar á meðal útgáfustýringu, samvinnudrifna greiningu og faglega framsetningu gagna og niðurstaðna. Reynsla úr námskeiðunum sýnir aukna þátttöku, sterkari teymisvinnu, minni laumufarþegavanda og rýmri skilning á fagmennsku í samskiptum við ytri samstarfsaðila. Kennarar fá jafnframt skýrari innsýn í einstaklingsframlög og verja minni tíma í lokaúrvinnslu. Samspil gagnsæis, ábyrgðar og raunhæfs vinnulags varpar þannig ljósi á hvernig kennsluaðferðir sem byggja á skýrum, endurteknum ferlum og traustri endurgjöf geta styrkt það sem telst góð háskólakennsla.
-
2024
Endurskoðun á námskeiði í Viðskiptagreind: Hagnýt hæfni í brennidepli
Ráðstefna Kennsluakademíu opinberu háskólanna, Veröld Háskóla Íslands, Reykjavík, 2024
Nothing to show at this detail level.
Theses
-
2016
ALICE: Analysis & Learning Iterative Consecutive Executions
Thesisp. 265, University of Iceland, 2016
-
2010
Nothing to show at this detail level.