DISA Seminar November 1st on Visualization Perspectives in Explainable AI

14 October, 2021
  • When? November 1st, 2021 at 12-13
  • Where? Online, links will be sent to those registered
  • Registration via this link

This talk with Professor Andreas Kerren, will overview interactive data visualization research with a focus on the development and use of visualization techniques for explainable artificial intelligence. The field of Information Visualization (InfoVis) uses interactive visualization techniques to help people understand and analyze data. It centers on abstract data without spatial correspondences; that is, usually it is not possible to map this information directly to the physical world. This data is typically inherently discrete. The related field of Visual Analytics (VA) focuses on the analytical reasoning of typically large and complex (often heterogeneous) data sets and combines techniques from interactive visualizations with computational analysis methods. I will show how these two fields belong together and highlight their potential to efficiently analyze data and Machine Learning (ML) models with diverse applications in the context of data-intensive sciences. As ML models are considered as complex and their internal operations are mostly hidden in black boxes, it becomes difficult for model developers but also for analysts to assess and trust their results. Moreover, choosing appropriate ML algorithms or setting hyperparameters are further challenges where the human in the loop is necessary. I will exemplify solutions of some of these challenges with the help of a selection of visualization showcases recently developed by my research groups. These visual analytics examples range from the visual exploration of the most performant and most diverse models for the creation of stacking ensembles (i.e., multiple classifier systems) to ideas of making the black boxes of complex dimensionality reduction techniques more transparent in order to increase the trust into their results.

Keywords:
information visualization, visual analytics, explainable AI, interaction, machine learning models, trust, explorative analysis, dimensionality reduction, high-dimensional data analysis

Further reading:
https://doi.org/10.1109/TVCG.2020.3030352
https://doi.org/10.1111/cgf.14034
https://doi.org/10.1109/TVCG.2020.2986996
https://doi.org/10.1111/cgf.14300
https://doi.org/10.1109/CSCS52396.2021.00008
https://doi.org/10.1177%2F1473871620904671

 

Seminar October 18th – Future Position X

1 October, 2021
  • When? Monday October 18th 12-13
  • Where? Online, link will be sent to those who sign up via this link https://forms.gle/NTo7jnysyLkBWaAm8 no later than October 15th

During the seminar Magnus Engström, CTO at Future Position X (FPX) will talk about two clear cases where FPX with data science has contributed to creating the conditions for a viable city center by collecting and combining data from different sources. More specifically, it will be about how we have applied machine learning to be able to predict movements in the city center and how we with a data-driven approach have created an application that helps the University of Gävle to conduct research on how Gävle residents experience their local environment. The presentation will be followed by a Q&A and discussion about potential collaborations with researchers from Linneaus University

Future Position X is an independent Swedish innovation center that works for growth through better health and well-being in the smart, sustainable and vibrant city. FPX contributes both technology and expertise to develop data-driven community solutions.

By initiating projects, creating relationships and building collaborations, FPX contributes to collaboration between business, academia and the public sector. FPX contributes to knowledge development of new technology by creating meeting places and networks around data-driven innovation such as GIS, AI, Internet of Things and blockchain technology. FPX also provides technical solutions, including the Innovation Platform, a data platform that can be used to digitally model societies. We are an important player in the work of strengthening both society and companies to a more sustainable growth.

 

 

Workshop “Critical perspectives on cultural heritage: Re-visiting digitisation” 26 October, 9-12hrs

28 September, 2021

Organizers: The workshop is co-organized by Linnaeus University (Centre for Applied Heritage and iInstitute) and Swedish National Heritage Board

Website: https://lnu.se/en/meet-linnaeus-university/current/events/2021/critical-perspectives-on-cultural-heritage-re-visiting-digitisation/

About: Today, the Semantic Web and Linked Open Data are creating new value for the descriptive information in the cultural heritage sector. Libraries, museums, heritage management and archives are seeing new possibilities in sharing by turning their catalogues into open datasets that can be directly accessed, allowing cultural heritage data to be circulated, navigated, analyzed and re-arranged at unprecedented levels. This is supported by research funding bodies, governments and EU policies and numerous political interests, resulting in enormous investment in digitization projects which make cultural heritage information openly available and machine readable. But before deploying this data, one must ask: is this data fit for deployment?

Libraries, museums, heritage management and archives have long histories. Both the collections they house and the language they use(d) to describe said collections are products of that historical legacy, shaped by, amongst others, institutionalized colonialism, racism and patriarchy. Yet descriptive information is now being digitized and shared as if that legacy is not inherent to the collections. Instead, existing units of information are being distributed through new Web 3.0 technologies, bringing with it an outdated knowledge-base. Besides the risk of progressive techniques being applied to regressive content, we may also sacrifice the development of new knowledge in libraries, museums, heritage management and archives aimed at facilitating socially sustainable futures, remediating exploitative historical legacies.

For this workshop, we have invited researchers and practitioners to discuss ways in which digitisation approaches may be set up to change the nature and legacy of cultural collection prior to digital dissemination.

Welcome!

DISA Seminar October 4th on Aggregation as Unsupervised Learning and some of its Applications

24 September, 2021
  • When? October 4th 12-13
  • Where? Online – the link will be sent to those who sign up
  • Registration? Sign up via this link no later than October 3rd.

This seminar will be presented by the DISTA research group within DISA, you will meet and listen to Welf Löwe, Maria Ulan, Morgan Ericsson, Anna Wingkvist

Aggregation combines several independent variables to a dependent variable. The independent variables are different, possible mutually dependent observations of a real world. The dependent variable should preserve properties of the independent variables, e.g., the ranking or relative distance of the independent variable tuples, and ultimately the properties of the real world. However, while there usually exist large amounts independent variable tuples, there is no ground truth data available mapping these tuples to the corresponding dependent variable values. This makes aggregation an unsupervised machine learning problem, as opposed to, e.g., regression where data comprises independent variable tuples and the corresponding dependent variable values.

Instances of the problem frequently occur in software engineering, e.g., when trying to assess the quality of software by metrics. Metrics (independent variables) can easily be measured for a lot of software artifacts, but it is hard to measure quality (dependent variable). Instances also occur in many other assessment situations including, but not limited to the assessment of project proposals, financial investments, and human movements.

In our talk, we present
1) aggregation as unsupervised learning including unweighted and weighted approaches
2) ways to evaluate and compare different aggregation approaches including an evaluation of the approaches introduced in 1)
3) applications to software engineering problems applying the evaluation introduce in 2)

The recording of this session and previous recordings will be available at the following link

NEW DISA Seminar Series starting September 6th 12-13

2 September, 2021

We are now finally starting a new Seminar series within DISA, even if you are not affiliated with DISA you are welcome to attend.

Aim with the seminar series:
Our research centre now have some 10 different research groups, each comprising a trending research topic. In order to make those different subjects of expertize more known outside of the own group and more accessible to PhD students we now launch a research seminar series.

Out first lunch seminar series will be on Monday September 6th 12-13 with Thomas Holgersson
Link to the seminar: https://lnu-se.zoom.us/j/63536937748 (no sign up needed)

Titel: Matrices in different dimensions: high, low and in between
Abstract: I will survey some common methods for statistical analysis of random matrices in fixed and in increasing dimensions. The geometry of high-dimensional objects will discussed from a data-analytic perspective. I will also cover some different modes of asymptotics, with particular focus on scalability.
Keywords: Wishart ensambles, geometry of high-dimensional objects, spectral analysis, Mahalanobis distance, modes of convergence.

Kind regards,
Thomas and Diana

Summer/winter schools available through EUniWell

20 May, 2021

Here is a overview of Summer/Winter schools at Leiden University (in particular at Leiden Medical Center) that is available to us through EUniWell, might be of interest for some of you.

  • Data Science in Health and Disease – Leiden-Brazil Summer School (June 2021, online)
  • Population Health Management – LUMC Summer School (June 2021, online)
  • TechMed – LUMC Summer School (July 2021, online)
  • Artificial Intelligence & Value Based Healthcare – LUMC Summer School (August 2021, online)
  • Regenerative Medicine – Leiden Summer School (October 2021)
  • Translational Research on Neuromuscular Diseases – LUMC / ERN EURO-NMD / TREAT-NMD Winter School (December 2021)

More information and registration:

https://www.universiteitleiden.nl/en/education/study-programmes/summer-schools?pageNumber=1&interest=medicine

iInstitute / Digital Humanities webinar: The Ethics of Datafication and AI by Geoffrey Rockwell

18 May, 2021

Summary – We all want artificial intelligence to be responsible, trustworthy, and good… the question is how to get beyond principles and check lists. In this paper I will argue for the importance of the data used in training machines, especially when it comes to avoiding bias. Further, I will argue that there is a role for humanists and others who have been concerned with the datafication of the cultural record for some time. Not only have we traditionally been concerned with social, political and ethical issues, but we have developed practices around the curation of the cultural record. We need to ask about the ethics around big data and the creation of training sets. We need to advocate for an ethic of care and repair when it comes to digital archives that can have cascading impact.

About the speaker – Geoffrey Rockwell is a Professor of Philosophy and Digital Humanities, Director of the Kule Institute for Advanced Study and Associate Director of AI for Society signature area at the University of Alberta. He publishes on textual visualization, text analysis, ethics of technology and on digital humanities including a co-authored book Hermeneutica from MIT Press (2016). He is co-developer of Voyant Tools (voyant-tools.org), an award winning suite of text analysis tools. He is currently the President of the Canadian Society for Digital Humanities.

INVITATION: Applications for Marie Skłodowska-Curie Actions MSCA Postdoctoral Fellowships (MSCA-PF) 2021

11 May, 2021

If you work within the research areas of data intensive research and are interested in spending one or two years as a researcher at the Linnaeus University Centre for Data Intensive Sciences and Applications (DISA, https://lnu.se/en/disa), you are welcome to submit a proposal for participating in a DISA effort. DISA researches open questions in collection, analysis and utilization of large data sets applied to thematic areas such as astrophysics, mechanical engineering, construction, eHealth, social sciences, and the humanities, see details below. With its core in computer science, it takes a multidisciplinary approach and collaborates with researchers from all faculties at the university.

The effort is supported by our Grants and Innovation Office (GIO) and should lead to an application for funding from MSCA-PF, allowing you to conduct a PostDoc research project and participate in the DISA activities.

We welcome all proposals within our core and thematic research fields, but are especially interested in applications that address data intensive research conducted in or together with one of our thematic areas.

A few important:

  • May 23th 2021 – Deadline for submitting initial proposals.
  • June 4th 2021 (a.m.) – Online MSCA-PF information. Presentation of important points and formalities.
  • June 27th 2021 – Deadline for submitting first drafts of full MSCA-PF applications
  • July 1st – Feedback session between supervisors and the applicant
  • Preliminary 15 September, 5 pm (Swedish time) – Deadline for submitting MSCA-PF applications to the European commission.

The European Marie Skłodowska-Curie Postdoctoral Fellowships are open to researchers moving within Europe, as well as those coming in from other parts of the world, as long as one has not lived or worked in Sweden for more than 12 months during the last 3 years (preliminary reference date: 15 September 2021). If the application is approved, you must move to Sweden. The applicant should be an experienced researcher, meaning that she or he should hold a doctoral degree (or receive such a degree before 9/2021). There is also an upper age limit: eligible researchers have a maximum of 8 years full-time equivalent experience in research, measured from the date of award of the doctoral degree. Years of experience outside research and career breaks (e.g., due to parental leave) are deductible, similarly years outside Europe (for European citizens and long-term residents).

Read the rest of this entry »

Welcome to iInstitute / DH seminar: On the “Art of Losing”—Some Notes on Digitization, Copying, and Cultural Heritage

16 February, 2021

When? 4 March, 9am
Location: https://lnu-se.zoom.us/j/64735625753

On the “Art of Losing”—Some Notes on Digitization, Copying, and Cultural Heritage
Copying is a creative “art of losing” that sustains culture and lends substance to heritage. This talk will aim to unpack the meaning of this statement and unravel some of the many paradoxes inherent in copying what has been inherited as culture using digital technologies. How is it that cultural reproduction and representation always entail loss while also always perpetuate things and ideas valued as culture and as heritage? What kinds of loss do digital technologies ensure? In what ways do new digital technologies sustain culture and enable heritage to persist? Attempting to unravel some of the conceptual and practical knots that formulate riddles like these, the first half of the talk will investigate a few key terms—copying, culture, and heritage. It will also survey a few of the important technologies used to copy texts in East Asia and on the Korean peninsula—brushes, bamboo slips, paper, woodblocks, new and old forms of movable metal type, photography and various forms of lithography, digital imaging, encoding schema, and forms of machine learning. This brief survey will help to situate a discussion in the second half of the talk about the current state of our creative “arts of loss” as they concern creating digital copies of cultural heritage objects. To suggest the current state of our arts, two research projects will be introduced. The first is an effort by nuns at the Taiwanese Buddhist Temple Fo Guang Shan to create an accurate digital transcription of every historical instantiation of the massive Buddhist canon. Their aim is to help ensure Buddhist heritage. The second is an effort by the National Library of Korea to make more of Korea’s textual heritage available to its patrons as digital transcriptions by using deep learning to overcome long-standing difficulties associated with the automated transcription of Korean texts. The American poet Elizabeth Bishop suggests in her poem “One Art” that “the art of losing is not hard to master.” This talk will suggest that Bishop’s poetic insight is helpful for thinking about digitization and cultural heritage. Loss is inevitable when reproducing cultural heritage by means of digital technologies. These losses are not necessarily a disaster. Each copy makes what has been inherited available again to new places and times. But how we practice this “art of losing” is important to consider since how we copy with our digital tools formulates what is inherited as cultural heritage.

About Wayne de Fremery , he is an associate professor in the School of Media, Arts, and Science at Sogang University in Seoul and Director of the Korea Text Initiative at the Cambridge Institute for the Study of Korea in Cambridge, Massachusetts (http://www.koreatext.org/). He also currently represents the Korean National Body at ISO as Convener of a working group on document description, processing languages, and semantic metadata (ISO/IEC JTC 1/SC 34 WG 9). Wayne’s research integrates approaches from literary studies, bibliography, and design, as well as information science and artificial intelligence. Recent articles and book chapters by Wayne have appeared in The Materiality of Reading (Aarhus University Press, 2020), The Wiley-Blackwell Companion to World Literature (Ken Seigneurie ed., 2020), and Library Hi-Tech (2020). Wayne’s bibliographical study of Chindallaekkot (Azaleas), a canonical book of modern Korean poetry, appeared in 2014 from Somyŏng Publishing. In 2011, his book-length translation of poetry by Jeongrye Choi, Instances, appeared from Parlor Press. Books designed and produced by Wayne have appeared from the Korea Institute at Harvard University, the University of Washington Press, and Tamal Vista Publications, an award-winning press he ran before joining the faculty of Sogang University. Some of his recent research projects have concerned the use of deep learning to improve Korean optical character recognition (funded by the National Library of Korea), technology and literary translation (paper forthcoming from Translation Review), and “copy theory” (paper under review). Wayne’s degrees are from Whitman College, Seoul National University, and Harvard.