A Distributed Multi-Sensor Machine Learning Approach to Earthquake Early Warning

Authors: Kevin Fauvel, Daniel Balouek-Thomert, Diego Melgar, Pedro Silva, Anthony Simonet, Gabriel Antoniu, Alexandru Costan, Véronique Masson, Manish Parashar, Ivan Rodero, Alexandre Termier403-411

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments show that DMSEEW is more accurate than the traditional seismometer-only approach and the combined-sensors (GPS and seismometers) approach that adopts the rule of relative strength.
Researcher Affiliation Academia 1Univ Rennes, Inria, CNRS, IRISA, Rennes, France 2Rutgers Discovery Informatics Institute, Rutgers University, New Jersey, USA 3Department of Earth Sciences, University of Oregon, Oregon, USA
Pseudocode No The paper describes the algorithm steps in text and provides a diagram (Figure 1), but it does not include a formal pseudocode block or an algorithm listing.
Open Source Code Yes In addition, we render public our real-world dataset collected and validated with geoscientists and we make public reference to the code of our machine learning algorithms used.
Open Datasets Yes We employ a real-world dataset1 (Fauvel et al. 2019) composed of GPS and seismometers data on normal activity/medium earthquakes/large earthquakes collected and validated with geoscientists. 1https://figshare.com/articles/Earthquake_Early_Warning_Dataset/9758555
Dataset Splits Yes We performed a stratified k-fold cross-validation which kept the same proportion of earthquakes of different categories for each fold. K is set to 3 considering the number of large earthquakes (14 earthquakes). We present the dataset split in Table 2.
Hardware Specification No The paper discusses the need for 'high-performance computing techniques and equipments' and 'well-provisioned computing systems' but does not specify any particular hardware components such as CPU or GPU models, or memory specifications used for the experiments.
Software Dependencies No The paper mentions several software packages like WEASEL+MUSE, MLSTM-FCN, scikit-learn, xgboost, and hyperopt, often referring to their public implementations. However, it generally does not provide specific version numbers for these libraries, which is necessary for full reproducibility.
Experiment Setup Yes WEASEL+MUSE: we use the public implementation5 with the recommended settings (SFA word lengths l in [2,4,6], windows length in [4:60], chi=2, bias=1, p=0.1, c=5 and a solver equals to L2R LR DUAL) (Sch afer and Leser 2017); MLSTM-FCN, we test the public implementation6 based on the original paper (Fazle, Majumdar, and Harford 2018), using the recommended settings (128-256-128 filters, 250 training epochs, a dropout of 0.8 and a batch size of 128); ... Hyperparameters of classifiers at central level are set by hyperopt, a sequential model-based optimization using a tree of Parzen estimators search algorithm (Bergstra, Yamins, and Cox 2013).