7. Release Plans

The following features are planned to be developed for upcoming releases of quoFEM app. We are actively working on the features in the next release. Farther development priorities may change depending on feedback from the community. If you have any suggestions, we encourage you to contribute and contact us through the SimCenter Forum.

7.1. Sept 2023

  1. Bayesian model selection and averaging (1.2.4.1) - Define weights for each model candidate and propagate epistemic uncertainty by randomly selecting a model for each realization according to these weights.

  2. New, state-of-the-art surrogate methods - Heteroscedastic Gaussian Process Surrogates (1.2.2.2) - First, train a GP surrogate to capture the mean response, then train another GP surrogate the capture the variance using the residuals. This approach is especially powerful when large differences are observed in the variance of the output in different parts of the input space.

  3. Efficient forward propagation using Multi-Fidelity Monte-Carlo (1.2.3.1) - Use a few realizations of the output calculated with expensive, high-fidelity models to evaluate and correct the bias in efficient, approximate models, such as surrogates. With little extra computational demands, this method provides a substantial improvement in surrogate model performance if you have the high-fidelity models available that were used to train the surrogates.

  4. Accelerate Bayesian calibration using advanced Markov Chain Monte Carlo methods (1.2.3.3) - Build a surrogate model iteratively that will replace the original model in MCMC sampling. In each iteration, use a batch of data generated by evaluating the original model at carefully selected points to improve the surrogate model approximation of the target probability density. This is especially helpful when the original model is expensive to run and conventional MCMC is not feasible.

  5. Additional calibration examples developed with researchers running projects at NHERI experimental facilities (1.2.5.3) - These examples demonstrate real-world application of quoFEM app and they can also serve as templates for future users of NHERI facilities.

7.2. March 2024

  1. Advanced model calibration methods using recursive stochastic filters and particle filters (1.2.3.4)

Note

The numbers in parentheses are for internal tracking purposes.