pnnl/SNAP

RESOURCE

Abstract

In this work, we detail two uncertainty quantification (UQ) methods that provide complementary information. Readout ensembling, by finetuning only the readout layers of an ensemble of foundation models, provides information about model uncertainty. Amending the final readout layer to predict upper and lower quantiles replaces point predictions with distributional predictions, which provide information about uncertainty within the underlying training data. We demonstrate our approach with the MACE-MP-0 model, applying UQ to both the foundation model and a series of finetuned models. The uncertainties produced by the ensemble and quantile methods are demonstrated to be distinct measures by which the quality of the NNP output can be judged.
Developers:
Pope, Jenna (Bilbrey) [1] Choudhury, Sutanay [1] Firoz, Jesun [1]
  1. Pacific Northwest National Laboratory
Release Date:
2025-03-25
Project Type:
Open Source, Publicly Available Repository
Software Type:
Scientific
Licenses:
BSD 2-clause "Simplified" License
Sponsoring Org.:
Code ID:
153158
Site Accession Number:
Battelle IPID 33288-E
Research Org.:
Pacific Northwest National Laboratory
Country of Origin:
United States

RESOURCE

Citation Formats

Pope, Jenna (Bilbrey), Choudhury, Sutanay, and Firoz, Jesun. pnnl/SNAP. Computer Software. https://github.com/pnnl/snap. US DOE. 25 Mar. 2025. Web. doi:10.11578/dc.20250325.2.
Pope, Jenna (Bilbrey), Choudhury, Sutanay, & Firoz, Jesun. (2025, March 25). pnnl/SNAP. [Computer software]. https://github.com/pnnl/snap. https://doi.org/10.11578/dc.20250325.2.
Pope, Jenna (Bilbrey), Choudhury, Sutanay, and Firoz, Jesun. "pnnl/SNAP." Computer software. March 25, 2025. https://github.com/pnnl/snap. https://doi.org/10.11578/dc.20250325.2.
@misc{ doecode_153158,
title = {pnnl/SNAP},
author = {Pope, Jenna (Bilbrey) and Choudhury, Sutanay and Firoz, Jesun},
abstractNote = {In this work, we detail two uncertainty quantification (UQ) methods that provide complementary information. Readout ensembling, by finetuning only the readout layers of an ensemble of foundation models, provides information about model uncertainty. Amending the final readout layer to predict upper and lower quantiles replaces point predictions with distributional predictions, which provide information about uncertainty within the underlying training data. We demonstrate our approach with the MACE-MP-0 model, applying UQ to both the foundation model and a series of finetuned models. The uncertainties produced by the ensemble and quantile methods are demonstrated to be distinct measures by which the quality of the NNP output can be judged.},
doi = {10.11578/dc.20250325.2},
url = {https://doi.org/10.11578/dc.20250325.2},
howpublished = {[Computer Software] \url{https://doi.org/10.11578/dc.20250325.2}},
year = {2025},
month = {mar}
}