ParallelAcquisition System

Overview

For performing parallel active learning, an acquisition function is needed for prioritizing under what input conditions the high-fidelity model needs to be run. Once the high-fidelity model evaluations are performed, these outputs serve as additional training data for re-training the probabilistic ML model (e.g., Gaussian processes). The acquisition function can be defined in the [ParallelAcquisition] block.

Creating an ParallelAcquisition function

An acquisition function is created by inheriting from ParallelAcquisitionFunctionBase and overriding the computeAcquisitionInternal method in the base class. See the BayesianPosteriorTargeted class for an example.

Example Input File Syntax

[Likelihood<<<{"href": "../Likelihood/index.html"}>>>]
  [gaussian]
    type = Gaussian<<<{"description": "Gaussian likelihood function evaluating the model goodness against experiments.", "href": "../../source/likelihoods/Gaussian.html"}>>>
    noise<<<{"description": "Experimental noise plus model deviations against experiments."}>>> = 'noise_specified/noise_specified'
    file_name<<<{"description": "Name of the CSV file with experimental values."}>>> = 'exp1.csv'
    log_likelihood<<<{"description": "Compute log-likelihood or likelihood."}>>> = true
  []
[]

Available Objects

• Stochastic Tools App
• BayesianPosteriorTargetedBayesian posterior targeted from El Gammal et al. 2023.
• CoefficientOfVariationCoefficient of variation acquisition function.
• ExpectedImprovementExpected improvement acquisition function.
• ExpectedImprovementGlobalFitExpected improvement for global fit (EIGF) by Lam and Notz 2008.
• ProbabilityofImprovementProbability of improvement acquisition function.
• UFunctionU-function acquisition function for rare events analysis.
• UpperConfidenceBoundUpper Confidence Bound acquisition function.

Available Actions

• Stochastic Tools App
• AddParallelAcquisitionActionAdds acquistion function objects.