Working Groups
The research tasks will be carried out by five WGs. High interaction among the researchers of all the WGs will be fostered. Each WG will have autonomy to develop their results, which will constitute a piece of the final outputs.
Task 1. Formalization of the problems.
Firstly, the working framework will be established taking the
Memorandum of Understanding (MoU) as starting point. Further discussions and the input of new participants will be considered. The priority lines will be defined in light of the practical demands. A close interaction between the end-users and experts who will develop the methods will be fostered. In this way the critical shortcomings of the state-of-the-art methods can be properly analyzed and the essential problems can be formalized. Some of the researchers of the Action already work in collaboration with end-users and this will facilitate the initial contacts. However, forums and face-to-face meetings will be organized to enlarge the vision of the Action and, as a consequence, its impact.
The aim is to develop methods tailored to specific real needs. Thus, the datasets to be analyzed have to be collected and be at disposal of the Action members in advance. To facilitate the storage and access to the data, open-source data platforms such as CKAN may be used. The datasets will be updated while the Action progresses.
Task 3. Models and methods: development of efficient solutions.
The next step is to investigate different robust strategies to approach computationally intensive key-problems such as robust clustering, decision learning trees, multiple testing and feature selection. Fast and numerically stable recursive estimation strategies will be developed. Expertise in HPC will be exploited to tackle those problems that cannot be considered using conventional sequential strategies.
Task 4. Resampling-based inference: development of efficient solutions.
Frequently the optimality of the traditional approaches relies on rigid distributional or model constraints that can be alleviated by using resampling-based methods. In these cases flexible, consistent and efficient resampling methods will be investigated. Given the computational burden of these kind of procedures, the strategies will be improved by using HPC.
Task 5. Implementation of the algorithms.
In order to make the results accessible and easy-to- apply by the end-users, specialized software will be realised according to standard rules to be defined by the network. The algorithms will be implemented in open-source packages. The packages will include graphical tools and a detailed documentation assisting in the use of the packages will be delivered.
Task 6: Applications and benchmarking.
Finally, the implemented procedures will be tested on the stored datasets. The practitioners will provide feedback so that the approaches are tuned if needed and best practice guides will be generated to assist end-users beyond the Action. Supporting results for data-based policies will be investigated, e.g. the determination of areas exposed to different environmental risks. As a final product, advanced and easy-to-use data-analysis tools ready to be exploited by stakeholders will be provided.
COST is supported by the EU Framework Programme Horizon 2020