Gonzalo Nápoles has been awarded a senior postdoc position by the FWO for his research proposal: “Hybrid and interpretable models based on Fuzzy Cognitive Maps for the modeling and simulation of complex systems.” (250kEU)
Abstract. Modeling and simulating complex systems allows experts in a given domain to investigate the effect of hypothetical scenarios before them to happen. Building such models is however inherently difficult since they heavily rely on the human knowledge which is difficult to extract and formalize, they either produce high simulation errors or perform as black boxes with little flexibility, they are difficult to control due to convergence issues, among other factors. This research will develop a new methodology termed neural cognitive modeling to cope with these problems. The new methodology relies on Fuzzy Cognitive Maps, which are a type of interpretable recurrent neural network with interesting properties when it comes to modeling complex systems. The main contributions include developing the design methodology to reason in presence of heterogeneous pieces of information, and new theorems that support the methodology and state what can we expect with the information we have at hand in terms of simulation error and stability. Moreover, the envisaged research will introduce theoretically sound learning algorithms to reduce the simulation error, improve the system convergence and optimize the network topology. Last but not least, an explicability module able to elucidate why the system produced a particular decision will be developed. Such explications will be generated in the form of natural language sentences and their confidence will be assessed by novel interpretability measures.