Emulating computer models with high-dimensional count output

Philosophical Transactions of the Royal Society A Mathematical Physical and Engineering Sciences, Journal Year: 2025, Volume and Issue: 383(2292)

Published: March 13, 2025

Computer models are used to study the real world, and often contain a large number of uncertain input parameters, produce outputs, may be expensive run need calibrating real-world observations useful for decision-making. Emulators as cheap surrogates simulator, trained on small simulations provide predictions with uncertainty at unseen inputs. In epidemiological applications, example compartmental or agent-based modelling spread infectious diseases, output is usually spatially temporally indexed, stochastic consists counts rather than continuous variables. Here, we consider emulating high-dimensional count from complex computer model using Poisson lognormal PCA (PLNPCA) emulator. We apply PLNPCA emulator fields COVID-19 England Wales compare this fitting emulators aggregations full output. show that performance generally comparable, while inherits desirable properties, including allowing predicted capturing correlations between providing samples representative true This article part theme issue ‘Uncertainty quantification healthcare biological systems (Part 1)’.

Language: Английский

---

Cross-validation-based sequential design for stochastic models

Philosophical Transactions of the Royal Society A Mathematical Physical and Engineering Sciences, Journal Year: 2025, Volume and Issue: 383(2293)

Published: April 2, 2025

Complex numerical models are increasingly being used in healthcare and epidemiology. To represent the complex features, modellers often make decision to include stochastic behaviour where repeated runs of model with identical inputs produce different outputs. When computational constraints limit number replications, heteroscedastic Gaussian processes can be as a fast surrogate, allowing for efficient emulation varying noise levels across input space. The accuracy any emulator is greatly dependent on design training data, sequential algorithms increase points iteratively based predefined criteria. For models, problem more challenging due possibility replicates at points. This article develops new method which scales well high-dimensional spaces. We build upon an existing deterministic using expected squared leave-one-out error criterion that balances exploration replication. compare our approach methods applying it agent-based COVID-19 model. Results demonstrate proposed performs noisy environments, offering scalable alternative methods. part theme issue ‘Uncertainty quantification biological systems (Part 2)’.

Language: Английский

---

Challenges in estimation, uncertainty quantification and elicitation for pandemic modelling

Epidemics, Journal Year: 2022, Volume and Issue: 38, P. 100547 - 100547

Published: Feb. 10, 2022

The estimation of parameters and model structure for informing infectious disease response has become a focal point the recent pandemic. However, it also highlighted plethora challenges remaining in fast robust extraction information using data models to help inform policy. In this paper, we identify discuss four broad paradigm relating modelling, namely Uncertainty Quantification framework, estimation, model-based inference prediction, expert judgement. We postulate priorities methodology facilitate preparation future pandemics.

Language: Английский

---

Development and Evaluation of Two Approaches of Visual Sensitivity Analysis to Support Epidemiological Modeling

IEEE Transactions on Visualization and Computer Graphics, Journal Year: 2022, Volume and Issue: unknown, P. 1 - 11

Published: Jan. 1, 2022

Computational modeling is a commonly used technology in many scientific disciplines and has played noticeable role combating the COVID-19 pandemic. Modeling scientists conduct sensitivity analysis frequently to observe monitor behavior of model during its development deployment. The traditional algorithmic ranking sensitivity different parameters usually does not provide with sufficient information understand interactions between outputs, while need large number runs order gain actionable for parameter optimization. To address above challenge, we developed compared two visual analytics approaches, namely: xmlns:xlink="http://www.w3.org/1999/xlink">algorithm-centric visualization-assisted , xmlns:xlink="http://www.w3.org/1999/xlink">visualization-centric algorithm-assisted . We evaluated approaches based on structured analysis tasks as well feedback domain experts. While work was carried out context epidemiological modeling, this are directly applicable variety processes featuring time series can be extended models other types outputs.

Language: Английский

---