Machine Learning as the Key to Optimization
Our PhD student Paula Harder focuses on the topics »Deep Learning« and »Climate Modeling« in cooperation with the University of Oxford. In her research work, she is developing, among other things, an emulator based on Artificial Intelligence that approximates the microphysics of the aerosol model and makes the calculations faster and more efficient. In computer technology, an emulator is a system that imitates another system in certain aspects. The goal is to enable climate predictions on a global scale, with very high precision and over long periods of time through Machine Learning. This represents an opportunity to, if not prevent, at least recognise and prepare for the consequences of climate change.
To accomplish this, eleven million input-output data pairs were first generated using the aerosol model. This data was then used to train a neural network to replace the costly origin model. Subsequently, additional computational constraints were incorporated to overcome physical constraints – such as conservation of mass and positivity – were taken into account.