The BAREFOOT Framework

Batch Reification/Fusion Optimization Framework

Mobirise
The BAREFOOT framework is the main output from my Ph.D. studies. This framework combines two optimization approaches into a combined approach that is capable of optimizing a function faster than traditional Bayesian Optimization approaches and also faster than either of the two individual approaches. The framework has already gone through several iterations to improve performance. 
The BAREFOOT Framework is available as open source Python Code (see below for Github Repository and Code Ocean Capsule links). The development of the code is ongoing and there are efforts to expand the capabilities of the framework to enable multi-objective optimization as well as additional methods for evaluating the Acquisition Functions.

A schematic showing how the framework operations are structure is shown in the Figure above. In this overview, the background colors are used to indicate different sections of the code. The green section involves queries to the true models and the construction of the Gaussian Processes, the light blue section is for the batch optimization related to the reduced order models, which the light purple section is for the batch optimization related to the Ground Truth. Using these two steps ensures that it is possible to reduce the time of the optimization. 

It was chosen to implement the BAREFOOT framework as a Python Class. This was done to ensure that the code is as modular as possible. Having this modular structure is a large advantage in terms of upgrading the functionality of the framework. For more details on the exact specifics of how the code operates, please see the references at the bottom of this page.

  1. Code Ocean Capsule - An initial version of the framework has been set up in a Code Ocean Capsule. This capsule contains a simple example code that shows how the framework operates.
  2. Github Repository - a Github Repository contains all the code including the latest updates to the code.
  3. NanoHUB - A tutorial on the approaches used in the BAREFOOT Framework was presented in the Tutorial Series on NanoHUB (YouTube). There is an implemented tool on NanoHUB that is available for downloading and testing (https://nanohub.org/tools/barefoot/)

Published Articles

1

Couperthwaite, Richard, Abhilash Molkeri, Danial Khatamsaz, Ankit Srivastava, Douglas Allaire, and Raymundo Arroyave. “Materials Design Through Batch Bayesian Optimization with Multisource Information Fusion.” JOM, October 13, 2020. https://doi.org/10.1007/s11837-020-04396-x.

2

Couperthwaite, R., Khatamsaz, D., Molkeri, A., James, J., Srivastava, A., Allaire, D., Arróyave, R., "The BAREFOOT Optimization Framework." Integrating Materials and Manufacturing Innovation, 2021. https://doi.org/10.1007/s40192-021-00235-2.

It is sometimes a mistake to climb; it is always a mistake never even to make the attempt.

If you do not climb, you will not fall.

This is true.

But is it that bad to fail, that hard to fall?
-Neil Gaiman, Fables & Reflections (Sandman #6)

This site was designed with Mobirise