Current Research
- Uncertainty Quantification and Validation
- Representing and learning the uncertainties in complex, physics-based computational models. However, the form of the computational model greatly influences our approach to quantifying uncertainties. If the model is defined by a system of equations, we explore learning the model-form error through embedded model corrections. If the model is defined by black-box observations, we explore a posterior representations of uncertainties.
- Building data-driven models that adhere to physical constraints
- Calibration and validation algorithms
Current Projects
- ACCurate and Reliable Uncertainty Estimates (ACCRUE)
- Generating probabilistic forecasts by augmenting black-box outputs with an uncertainty distribution.
- Extending to non-Guassian distributions.
- Model-form error in ODEs with missing variables
- Model-form error in systems with missing variables and simplified physics
Previous Projects
- Validating Neural-Network-Corrected Dynamical Systems
- Calibrated and validated a neural-network-corrected compartmental disease model.
- Investigated how calibration data and the neural network architecture affected the validation time horizon.
- Global Optimization of Chemical Systems
- Investigated Basin-Hopping and Minima Hopping combinations and modifications in order to accuratly and efficiently predict the ground state of Lennard-Jones systems
- Software for Global Optimization
- Modeling Microstructure Evolution during Additive Manufacturing
- Employed a data science approach and machine learning to simulated additive manufacturing.