Metamodeling (also known as surrogate modelling) has become an advanced modelling technique that is applied in many fields of engineering. In a sense, a metamodel constructs a model of a model. Let us consider an example to better explain how it functions. Suppose you want to model an RF amplifier. This amplifier’s behaviour depends on a lot of different external settings: the input power, the frequency bandwidth of the excitation, supply voltage, etc. For each fixed settings combination, the RF amplifier will exhibit a certain dynamic behaviour. The goal of metamodeling is to model this dynamic behaviour as a function of the settings combinations.
Most of the existing metamodeling approaches use a global approach to build the metamodel (see figure). In the global setting, only one model is needed to describe the whole system. Unfortunately, due to the increasing amount of settings in modern systems (think about 5G for example), this global approach quickly becomes very computationally demanding. In some cases, it becomes even impossible to estimate the metamodel. To counteract this so-called curse of dimensionality we would like to investigate a novel approach based on local modelling techniques. In the local approach, we use multiple local models to describe the global behaviour of the system.
During this master thesis you will
Study and compare existing metamodeling and local modelling techniques.
Extend the local modelling techniques to handle multidimensional variations of the settings.
Apply your developed technique to challenging applications in electronics and photonics.
Interested or curious to know more? Feel free to contact Dries Peumans
Francesco Ferranti (B-PHOT, VUB)