guest ::
| Home > In process > Comparison of Epistemic Uncertainty Quantification Methods for Out-of-Distribution Detection in Autoencoder–RNN Surrogate Model of Molecular-Continuum Flow Simulations |
| Home > In process > Comparison of Epistemic Uncertainty Quantification Methods for Out-of-Distribution Detection in Autoencoder–RNN Surrogate Model of Molecular-Continuum Flow Simulations |
| Journal Article/Contribution to a conference proceedings/Contribution to a book | PUBDB-2026-02005 |
; ; ; ; ; ;
2026
Springer
Heidelberg
ISBN: 978-3-032-29918-5, 978-3-032-29918-5 (electronic)
This record in other databases:
Please use a persistent id in citations: doi:https://doi.org/10.1007/978-3-032-29918-5_31 doi:10.1007/978-3-032-29918-5_31
Abstract: Neural network surrogates have been shown to decrease computational costs of simulations, but often at the risk of unreliable predictions. This work integrates and evaluates multiple epistemic uncertainty quantification methods for a reproduced convolutional autoencoder–recurrent neural network surrogate architecture for molecular data in a coupled spatiotemporal molecular-continuum flow prediction. The surrogate is trained on an idealized Kármán vortex street dataset generated using the molecular–continuum simulation framework MaMiCo, and evaluated on three out-of-distribution datasets with progressively increasing shifts from the training distributions. In the Autoencoder model, the Deep Ensemble method sets a strong baseline, but after fine-tuning, both Gaussian Processes and Evidential Deep Learning show promising detection skills and faster inference than Deep Ensemble. This trend continues in the Autoencoder-RNN, which employs a propagation approach for Evidential distributions and an RNN-influenced latent space for Gaussian Processes.
; SCOPUS
|
The record appears in these collections: |
ZIP