‘Hybrid Prediction’ is a term in data science that reflects the integration of all sources of information and knowledge from a domain problem into an analytics framework to produce more accurate predictions a and methods for uncertainty quantification that translate into more confident and pragmatic decision-making.
Information and knowledge can be characterised by a physical model of the processes but these can be slow to run and complicated to represent and communicate.
Emulators are an approach that attempt to mimic a physical model using machine learning (ML). ML and in particular neural networks are good at capturing complex relationships and dynamical representations that form the physical nature of these domain problems. The concept involves constructing a ML model from the mapping of the inputs to the outputs from the model through carefully constructed simulations that attempt to span the parameter space of likely input combinations.
A robust emulator has the capacity to be used in an inference setting to form predictions for specific ‘what-if’ scenarios and quantify the uncertainty of the predictions to inform decision-makers.
To find out more: https://research.csiro.au/mlai-fsp/activities/hybrid-prediction/