Dataset for "DeepONet-Accelerated Bayesian Inversion for Moving Boundary Problems"

Read full text Summarise with VIRAT
Authors: Marco Iglesias, Michael E. Causon, M. Matveev, A. Endruweit, Michael V. Tretyakov
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This dataset accompanies the manuscript “DeepONet-Accelerated Bayesian Inversion for Moving Boundary Problems”(arXiv preprint). It provides data and model artifacts used to train and evaluate Deep Operator Networks (DeepONet) for Bayesian inversion in moving boundary problems, with application to resin infusion. Contents 1. Input–output datasets input_data_batch_seed_2.h5, output_data_batch_seed_2.h5Testing dataset consisting of 10,000 input–output pairs generated using a resin infusion simulator. The format is identical to the training data.(Full training datasets are not included due to size constraints; seeds and generation code are available on GitHub.) 2. Prior and posterior ensembles prior_ensemble_100_NEn5000_seed91882.h5 posterior_ensemble_100_NEn5000_seed91882.h5 posterior_ensemble_for_real_NEn5000_seed91882.h5 These ensembles are generated using the full-order model (MATLAB implementation provided in the GitHub repository). They are included to enable direct comparison with surrogate-based inversion methods without requiring re-running the full inversion procedure.The prior ensemble is also used for DeepONet-EKI. 3. Trained models and inversion results output_NS40000_D400.tar.gzContains the best-performing DeepONet model trained on 40,000 samples, including: model weights test metrics uncertainty quantification outputs (mean and covariance) posterior ensembles from DeepONet-EKI for both synthetic and real data output_NS20000_D400.tar.gz, output_NS10000_D400.tar.gzSimilar contents for models trained on 20,000 and 10,000 samples, respectively. These are used to reproduce results comparing DeepONet-EKI performance across different training dataset sizes (synthetic cases). Here, D denotes the latent dimension as defined in the manuscript and GitHub repository. Reproducibility Code for data generation, model training, and inversion is available at: GitHub (licensed under the MIT License). This dataset is released under the Creative Commons Attribution 4.0 International License.

Fields of Study & Subjects

Model Reduction and Neural Networks Gaussian Processes and Bayesian Inference Advanced Multi-Objective Optimization Algorithms

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
1 January 2026
Language
EN