Accurate predictions of future weather conditions are essential for the safety of people and goods, and for the management of a wide range of economic activities. However, the precise prediction of high-impact local phenomena remains difficult and computationally expensive. The progress of Artificial Intelligence (AI) technologies, with early impressive applications to weather forecasting, offers unexpected opportunities for a new generation of weather predictions based on hybridisation of traditional physical models and AI, allowing for increased accuracy and timeliness in a cost-effective way.
Beyond the methodological and technological breakthroughs of these new developments, a particular emphasis is put on the development of explainable [Tiddi and Schlobach, 2022] and robust AI solutions, as well as on their knowledge transfer (sharing) and acceptance by the end-users [Kelly et. al., 2023]. However, there are several shortcomings as the whole process is not enough formalised and annotated.
It becomes crucial to enrich the Machine Learning-oriented Numerical Weather Prediction (NWP-ML) outputs with knowledge on their construction, enabling users to understand their meaning and context. These transfers of knowledge (sharing and formalization) adapted to the usage will allow end users, such as forecasters or outside professionals who need precise meteorological knowledge to act in their field, to accept their usages. Furthermore, part of this formalization (using machine and humain readable languages) [Trojahn et al., 2022, Giraldo et. al., 2017]contributes to promote reproducibility by providing an intelligible representation of the AI processes, which will be the basis for their comparison and reproducibility [Euzenat, 2022,Werner et. al., 2024]. Finally, this knowledge forms a part of the raw material for sociological studies to judge the acceptability levels of NWP-ML systems or the complements needed for their acceptance by professional groups. This thesis will focus on knowledge transfer (including reproducibility), what will serve as a basis for the sociological studies to characterise acceptability of AI-based solutions.
The objectives of the thesis are twofold:
With respect to (i), this will involve reusing, integrating, and extending semantic models in order to be able to represent such different kinds of knowledge (training sets provenance and quality, specific techniques, explainability outputs, uncertainty associated to the models, specific fine-tuning strategies applied, background knowledge on the type of prediction, non-experimented conditions, etc.) and guarantee their constraints in terms of interpretation. This intelligible representation of the AI process will be the basis for their comparison and reproducibility of AI results.
This work is part of an ambitious program within the ANITI EXPLEARTH chair on explainable and physics-informed AI for regional weather prediction.
References:
[Euzenat, 2022] Jérôme Euzenat, Beyond reproduction, experiments want to be understood, in: Proc. 2nd workshop on Scientific knowledge: representation, discovery, and assessment (SciK), Lyon (FR), pp774-778, 2022 https://moex.inria.fr/files/papers/euzenat2022a.pdf
[Giraldo et. al., 2017] Olga Giraldo, Alexander Garcia, Federico Lopez, and Oscar Corcho. 2017. Using semantics for representing experimental protocols. Journal of Biomedical Semantics 8 (2017), 52. doi:10.1186/s13326-017-0160-y
[Kelly et. al., 2023] Kelly, S., Kaye, S., Oviedo-Trespalacios, O., 2023: What factors contribute to the acceptance of artificial intelligence? A systematic review, Telematics and Informatics, Volume 77, 2023, 101925, ISSN 0736-5853, doi:10.1016/j.tele.2022.101925.
[Tiddi and Schlobach, 2022] Ilaria Tiddi, Stephan Schlobach, Knowledge graphs as tools for explainable machine learning: A survey, Artificial intelligence 302:103627, 2022 doi:10.1016/j.artint.2021.103627
[Trojahn et al., 2022] Trojahn, C., Kamel, M., Annane, A., Aussenac-Gilles, N., Baehr, C., FAIRification of Multidimensional and Tabular Data by Instantiating a Core Semantic Model with Domain Knowledge: Case of Meteorology, Proc. 16th International Conference on Metadata and Semantics Research, London. Communications in Computer and Information Science 1537, 2022
[Werner et. al., 2024] Luisa Werner, Pierre Genevès, Nabil Layaïda, Jérôme Euzenat, Damien Graux, Reproduce, replicate, reevaluate: the long but safe way to extend machine learning methods, in: Proc. 38th AAAI Conference on Artificial Intelligence (AAAI), Vancouver (CA), pp15850-15858, 2024 https://moex.inria.fr/files/papers/werner2024a.pdf
Links:
Qualification: Master or equivalent in computer science.
Researched skills:
Doctoral school: MSTII, Université Grenoble Alpes.
Advisor: Cássia Trojahn dos Santos (Cassia:Trojahn-dos-Santos#univ-grenoble-alpes.fr) and Jérôme David (Jerome:David#inria:fr)
Group: The successful candidate will join the mOeX team common to INRIA & LIG, with collaborations with the CNRM (Centre National de Recherches Météorologiques) and Meteo-France in Toulouse. mOeX is dedicated to study knowledge evolution through adaptation. It gathers researchers which have taken an active part these past 20 years in the development of the semantic web and more specifically ontology matching and data interlinking.
Place of work: The position is located at INRIA Montbonnot (near Grenoble) a main computer science research lab, in a stimulating research environment.
Hiring date: October 2025.
Duration: 36 months
Deadline: as soon as possible.
Contact: For further information, contact us.
Procedure: Interested candidates should contact Cassia Trojahn and Jérôme David, with a CV, cover letter, and references. They will also have to apply to Proposal 66130
File: Provide Vitæ, motivation letter and references. It is very good if you can provide a Master report and we will ask for your marks in Master, so if you have them, you can join them.