TKThierry Klein

Thierry Klein

*This is an algorithmically generated author record based on publicly available information.

Thierry Klein has a total of 147 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

    • 26 June 2023
    • Published by HAL in HAL
    Open Access

    A Multivariate Functional Data Analysis of Aircraft Trajectories

    by Remi Perrichon, Xavier Gendre, Thierry Klein, Rémi Perrichon

    While advanced methods for functional data analysis have recently been developed in the literature, applications to aircraft trajectories have remained scarce, despite operational relevance. One reason is the practical difficulties affiliated with the multivariate nature of trajectories and associated physical constraints. Indeed, an aircraft trajectory usually involves three dimensions in space (longitude, latitude, altitude) but also weather values (say wind speed and direction), each dimension having its specificities. To name a few, smoothing altitude values requires to ensure both non-negativity and boundary constraints. Wind directions have support on the unit circle. Additional to constrained smoothing challenges, phase variations are to be taken into account as flights are never of the same duration. To tackle these issues, two smoothing methods respectively based on constrained splines and asymmetric kernels are implemented on real data. For each approach, two strategies to handle the circular nature of wind directions are compared. Registration is performed. A joint pointwise test is proposed to demonstrate that delayed flights have experienced less favorable wind conditions.

  • Top Cited Papers
    Open Access

    Testing the Effect of Relative Pollen Productivity on the REVEALS Model: A Validated Reconstruction of Europe-Wide Holocene Vegetation

    by M. A. Serge, F. Mazier, R. Fyfe, M.-J. Gaillard, T. Klein, A. Lagnoux, D. Galop, E. Githumbi, M. Mindrescu, A. B. Nielsen, et al.

    Reliable quantitative vegetation reconstructions for Europe during the Holocene are crucial to improving our understanding of landscape dynamics, making it possible to assess the past effects of environmental variables and land-use change on ecosystems and biodiversity, and mitigating their effects in the future. We present here the most spatially extensive and temporally continuous pollen-based reconstructions of plant cover in Europe (at a spatial resolution of 1° × 1°) over the Holocene (last 11.7 ka BP) using the ‘Regional Estimates of VEgetation Abundance from Large Sites’ (REVEALS) model. This study has three main aims. First, to present the most accurate and reliable generation of REVEALS reconstructions across Europe so far. This has been achieved by including a larger number of pollen records compared to former analyses, in particular from the Mediterranean area. Second, to discuss methodological issues in the quantification of past land cover by using alternative datasets of relative pollen productivities (RPPs), one of the key input parameters of REVEALS, to test model sensitivity. Finally, to validate our reconstructions with the global forest change dataset. The results suggest that the RPPs.st1 (31 taxa) dataset is best suited to producing regional vegetation cover estimates for Europe. These reconstructions offer a long-term perspective providing unique possibilities to explore spatial-temporal changes in past land cover and biodiversity.


  • Open Access

    ASRS-CMFS vs. RoBERTa: Comparing Two Pre-Trained Language Models to Predict Anomalies in Aviation Occurrence Reports with a Low Volume of In-Domain Data Available

    by Samuel Kierszbaum, Thierry Klein, Laurent Lapasset

    We consider the problem of solving Natural Language Understanding (NLU) tasks characterized by domain-specific data. An effective approach consists of pre-training Transformer-based language models from scratch using domain-specific data before fine-tuning them on the task at hand. A low domain-specific data volume is problematic in this context, given that the performance of language models relies heavily on the abundance of data during pre-training. To study this problem, we create a benchmark replicating realistic field use of language models to classify aviation occurrences extracted from the Aviation Safety Reporting System (ASRS) corpus. We compare two language models on this new benchmark: ASRS-CMFS, a compact model inspired from RoBERTa, pre-trained from scratch using only little domain-specific data, and the regular RoBERTa model, with no domain-specific pre-training. The RoBERTa model benefits from its size advantage, while the ASRS-CMFS benefits from the pre-training from scratch strategy. We find no compelling statistical evidence that RoBERTa outperforms ASRS-CMFS, but we show that ASRS-CMFS is more compute-efficient than RoBERTa. We suggest that pre-training a compact model from scratch is a good strategy for solving domain-specific NLU tasks using Transformer-based language models in the context of domain-specific data scarcity.


  • Large Deviations at the Transition for Sums of Weibull-Like Random Variables

    by Fabien Brosset, Thierry Klein, Agnès Lagnoux, Pierre Petit


  • Open Access

    Modeling uncertainties in complex systems

    by Nabil Rachdi, Jean-Claude Fort, Thierry Klein, Fabien Mangeant