SEAM Surrogate Modeling and Design Optimization Course Material

Overview

This repository contains materials for an introductory course on "Surrogate Modeling and Design Optimization," originally developed for the Shared Education in Artificial intelligence and Machine learning (SEAM) professional development program at Lawrence Livermore National Laboratory (LLNL). It features the surmod Python package, which implements core techniques including Feedforward Neural Networks (FFNNs) and Gaussian Processes (GPs) as surrogates, basic Bayesian Optimization (BO) for design optimization, and Sensitivity Analysis (SA). Weekly workflow scripts in the scripts directory demonstrate these methods in practice. Example datasets included in the data directory are used to support hands-on learning and experimentation.

For more detailed documentation on the modules in the surmod package, please view our documentation.

Information about the SEAM program is available to employees of Lawrence Livermore National Laboratory at https://seam.llnl.gov.

Please send correspondence about this code to seam@llnl.gov.

Getting Started

1. Clone the repository

    git clone https://github.com/LLNL/SEAMsurrogates.git
    cd SEAMsurrogates

2. Create a virtual environment

    python3 -m venv .venv

3. Activate the virtual environment

    source .venv/bin/activate

5. Install the surmod package and dependencies locally

    pip install -e .

Datasets

The datasets in the data directory are from different sources:

• The JAG ICF dataset is from this repo.
• The borehole dataset was simulated from the borehole function.
• The datasets that start with hst are from simulations of the Hubble Space Telescope (HST), and are sampled from larger datasets available here.

Course Outline

Week	Topic	Driver Scripts
1	Intro to Deep Learning, PyTorch, and Surrogate Modeling
2	Intro to Neural Networks (NNs) as Surrogates	nn_sandbox.py
3	Exploring NN Surrogates for Example Data	nn_fromdata.py
4	Intro to Gaussian Processes (GPs) as Surrogates	gp_sandbox.py
5	Exploring GP Surrogates for Example Data	gp_fromdata.py
6	Basic Bayesian Optimization (BO) for Design Optimization	bo_sandbox.py
7	Exploring BO for Design Optimization on Example Data	bo_fromdata.py
8	Intro to Sensitivity Analysis (SA)	sa_sandbox.py, sa_fromdata.py
9	Discuss Problem Statement Ideas for Final Project
10	Work on Final Project
11	Work on Final Project
12	Final Project Review & Presentation

Contributors and Contact Information

• Jason Bernstein (bernstein8@llnl.gov)
• Emily Bogle (bogle4@llnl.gov)
• Andrew Gillette (gillette7@llnl.gov)
• Kevin Quinlan (quinlan5@llnl.gov)

License

This software is distributed under the terms of the BSD-Commercial license.

See LICENSE and NOTICE for details.

Release

LLNL-CODE-2010455