Documentation for the active learning (AL) workflow developed as a part of the article Sivaraman, G.; Jackson, N.E. "Coarse-Grained Density Functional Theory Predictions via Deep Kernel Learning" J. Chem. Theory Comput. 2022, 18, 2, 1129-1141. For more details, please refer to the paper.

The workflow is built as a PyTorch based GPU accelerated framework and offers the following capabilities:

• GPU accelerated Scalable Gaussian Processes and Exact Deep Kernel Learning (DKL) based on GPyTorch library
• Bayesian Optimization for hyperparameter tuning of DKL using the GPyOpt library
• PyTorch based numeric implementation of AL query strategies using standard GPR based uncertainty and beyond.
• Capable of running on the state-of-the-art NVIDIA A100 GPU's available at the LCRC SWING cluster and the ALCF ThetaGPU.

• Gaussian Process Regression: Single and Multi-task
• Deep Kernel Learning

• Matern
• RBF

• Random
• Uncertainty
• Expected Model Output Changes (EMOC)

Running GPyTorch on A100 GPU has the following basic requirment:

• MAGMA + CUDA 11.0

The step-by-step compilation is covered in INSTALLATION.MD

Running the workflow is extensively covered in RUN.MD

If you are using this active learning workflow in your research paper, please cite us as

@article{doi:10.1021/acs.jctc.1c01001,
author = {Sivaraman, Ganesh and Jackson, Nicholas E.},
title = {Coarse-Grained Density Functional Theory Predictions via Deep Kernel Learning},
journal = {Journal of Chemical Theory and Computation},
volume = {18},
number = {2},
pages = {1129-1141},
year = {2022},
doi = {10.1021/acs.jctc.1c01001},
    note ={PMID: 35020388},

URL = { 
        https://doi.org/10.1021/acs.jctc.1c01001
    
},
eprint = { 
        https://doi.org/10.1021/acs.jctc.1c01001
    
}

}

This material is based upon work supported by Laboratory Directed Research and Development (LDRD-CLS-1-630) funding from Argonne National Laboratory, provided by the Director, Office of Science, of the U.S. Department of Energy under Contract No. DE-AC02-06CH11357. N.E.J acknowledges support from the Dreyfus Program for Machine Learning in the Chemical Sciences and Engineering during this project. This research used resources of the Argonne Leadership Computing Facility, which is a DOE Office of Science User Facility supported under Contract DE-AC02-06CH11357. Argonne National Laboratory’s work was supported by the U.S. Department of Energy, Office of Science, under contract DE-AC02-06CH11357. We gratefully acknowledge the computing resources provided on Bebop and Swing, high-performance computing clusters operated by the Laboratory Computing Resource Center at Argonne National Laboratory.

• Ganesh Sivaraman gsivaraman:AT::anl.gov
• Nicholas Jackson jackson.nick.e::aT:gmail.com