The Batch Reification-Fusion Optimization framework, nicknamed BAREFOOT, combines two optimization approaches that have both been proven to enable accelerated materials design.

• Reification-Fusion: The first technique is the Reification-Fusion approach which is a multi-fidelity fusion approach. The aim of the Reification-Fusion approach is to fuse models to varying fidelity (accuracy) by first correcting the model errors to account for the accuracy of the model and then using the reification approach to estimate the correlation between each pair of models used. This enables the models to be fused using fusion under known correlation approaches. This approach creates a fused model that is more accurate with respect to the Ground Truth than any of the models used. Using this optimization approach, it is possible to decrease the number of Ground Truth queries and so has the potential to drastically reduce the cost of the optimization.
• Batch Bayesian Optimization: The second technique is a Batch Bayesian Optimization approach which uses a simple premise. This is that it is normally not practical or normally possible to determine the true hyperparameters for the surrogate models used to approximate the black box function that needs to be optimized. Under this premise, the technique samples hyperparameters from a given distribution and constructs a surrogate model for each set of hyperparameters. Each of these different surrogates will have different estimates for the next-best point to query. Using the k-medoids clustering technique these results are clustered into the number of clusters required by the batch size.

The aim of combining these approaches is to enable accelerated material design which can leverage the vast array of material property models that have been developed over the years. These range from simple, relatively inaccurate, empirical models to highly accurate finite element, DFT, or phase-field models. The nature of the framework also allows for direct integration with experimental approaches. Since there has been much development in the field of high-throughput computational and experimental approaches, a framework such as this can be used to guide these experiments.

This tool provides three Notebooks that demonstrate and compare the two techniques mentioned above as well as the full BAREFOOT Framework using a sample mechanical model problem.

DEMS Project Group (Texas A&M University)

Richard Couperthwaite

Dr. Raymundo Arroyave

Dr. Douglas Allaire

Dr. Ankit Srivastava

Abhilash Molkeri

Danial Khatamsaz

Jaylen James

The authors acknowledge grant No. NSF-CMMI-1663130, DEMS: Multi-Information Source Value of Information Based Design of Multiphase Structural Materials as well as grant No. NSF-S&AS-1849085. Raymundo Arroyave would also like to acknowledge grant No. NSF-1835690. Douglas Allaire would also like to acknowledge grant No. FA9550-15-1-0038.  Douglas Allaire and Raymundo Arroyave acknowledge grant No. NSF-DGE-1545403. Portions of this research were conducted with the advanced computing resources provided by Texas A&M High-Performance Research Computing Facility.

Couperthwaite, R., Molkeri, A., Khatamsaz, D. et al. Materials Design Through Batch Bayesian Optimization with Multisource Information Fusion. JOM 72, 4431–4443 (2020). https://doi.org/10.1007/s11837-020-04396-x

Researchers should cite this work as follows:

• Richard Couperthwaite, Raymundo Arroyave (2021), "A Batch Reification/Fusion Optimization Framework for Bayesian-based Material Optimization," https://nanohub.org/resources/barefoot. (DOI: 10.21981/EABV-2J50).

  BibTex | EndNote

1. BAREFOOT framework
2. batch processing
3. Bayesian
4. Bayesian statistics
5. k-methods clustering
6. Multifidelity
7. NCN Group - Machine Learning Hands-on
8. Optimization
9. python