Program Dag 2

10:30

How a metadata for machines workshop can help to improve research impact.

Nationally and internationally the FAIR principles must make way for valuable changes of research practice. DeiC has developed a ‘rapid M4M’ concept, which can be held in 2 x ½ days and gives one research group the unique opportunity to get a good understanding of the necessary first steps that have to be taken in order to produce machine actionable metadata. We will introduce the concept, and attempt to engage you in cooperation on how machine actionable metadata can be produced.

The benefit of machine-actionable metadata is that they become easier to discover and access both metadata and data, and the number of citable research outputs increases, while complying with the FAIR principles. By doing this the researcher will increase the impact of the research group's data, strengthen her/his data handling internally as well as within the peer group. Additionally, good data handling practices strongly increase chances of receiving funding for your research.

Our M4M concept has caught international attention as a way to create obvious change for research practices.

11:00

Rescuing legacy data from obsoletion: WindData.com use-case

Research groups often possess a rich heritage of data spanning periods of several decades. However, this legacy almost never gets proper attention due to lack of funding, and thus lack of long-term maintenance plans. As a consequence, the legacy data are typically unreadable and unaccessible due to obsolete formats and technologies in which they are provided. To prevent this from happening, DTU Wind Energy executed a pilot data rescue project and FAIRified the WindData.com data collection, moving it from an old to a new technology stack. In this talk, we will present the results of this work.

11:30

Making sensitive data FAIR through distributed learning

Data collected for research should ideally be available for a future research project and not just be discarded after a given project has been finalized. This idea is in line with the FAIR principles that data should be: findable, accessible, interoperable, and reusable. Within health science, it can be challenging to make data openly available due to the sensitive nature of patient data. The FAIR principles, therefore, need to be slightly modified within the area of health sciences. The presentation is related to the development of multi-centre data models within the field of sensitive data that prevent the pooling of all the data. A distributed learning method, where all data are kept locally and only aggregated data are collected, will be discussed as an alternative to having full access to all the data. This method can support the efficient reuse of data while protecting the sensitive nature of the data.

11:30

Exploring past economies with HPC-enabled agent-based modelling

Imagine going to a market to buy a cooking pot. The seller offers you a wide selection of locally made or imported ceramics, some cheaper, some more expensive. But which one to choose?

Here we present a model of economic interactions designed to investigate how simple customer preferences can shape centuries long term economic and cultural trends. By applying a number of standard cultural evolution algorithms to a baseline economic model we investigate how cultural behavioural scenarios can lead to different economic patterns in archaeological data. Did traders communicate with each other and copy strategies? And if so, how often and who did they copy from?

We test three competing hypotheses formalised in an agent-based model (ABM) and compare their predictions to archaeologically observed tableware distribution patterns using Approximate Bayesian Computation (ABC). Both techniques: ABM and ABC are highly computationally expensive. ABM due to the use of hundreds of independent autonomous agents, each performing their own calculations and ABC because it requires hundreds of thousands simulation runs to sweep through the parameter space and converge on the ones consistent with the target data pattern. To achieve this we deployed a custom solution developed at the Barcelona Supercomputing Center: an ABM framework, Pandora, which integrates ABM models with High Performance Computing environments.

13:30

The missing link in the knowledge ecosystem: publishing reproducible science

Knowledge Exchange, which is a knowledge forum with members from UK, Holland, Germany, Finland, France and Denmark, has initiated a project about “Publishing Reproducible Research Output”. The overall aim is to investigate current practices and barriers related to publishing reproducible research output and to determine how infrastructures (technical and social) can support this.

The rationale for the project is that despite the Open Science agenda puts more focus on transparency with regard to research processes, efforts to enhance reproducibility are still not part of everyday practice. There have been efforts to enhance reproducibility and transparency through journals offering more space for methods sections, checklists or reporting summaries/guidelines for authors, employing statisticians that check the calculations and offering or requiring authors to submit the data and code on which the publication is based. In addition, there are more and more journals that publish negative (results that do not support initial hypotheses) or null results (studies that have no effect). The talk will present the context and background for the project and results from the final report.