Tutorials

Educational platforms for Cyber-physical Systems and Internet of Things

Experiences in delivering lectures in Cyber-physical Systems and the Internet of Things will be shared. Emphasis will be given on providing affordable and timely hardware setups to study this novel and rapidly changing field. Overview of the hardware components will be given from microcontrollers, Mini PCs, sensors, actuators and controllers. Availability of the components on the market will be discussed. Software setup will be considered including operating systems, programming languages and internet/cloud integration. Syllabus will be presented as well as feedback from students. Interconnection between practical realization of advanced control systems and theory will be discussed. The future development in the field of CPS&IoT education will be discussed.

Integrating Prior Knowledge in Symbolic Regression

Many engineering domains deal with highly complex processes or systems. Modeling these systems requires significant effort and associated costs of observation in order to understand their underlying principles and produce a mathematical description of the system. Sometimes, this approach may be infeasible due to the complexity of the task, computational limits or high-dimensional nature of a system. Even if the underlying principles are fully known, a data-based modeling approach such as Symbolic Regression using Genetic Programming can be computationally more efficient and useful for approximating properties of a system. However, by contrast to first-principle models, data-based models have no guarantee that the model obeys physical laws, has an output within a desired boundary, or satisfies other desired properties. It is important in this context to include prior knowledge in order to find models that are consistent with the expected behavior of the studied system. The tutorial will deal with methods of integrating prior knowledge in form of intervals constraints on model output and its derivatives with respect to the input variables. We demonstrate how such intervals can be calculated with the help of automatic differentiation and interval arithmetic. We show preliminary results using this approach on a collection of physics problems from the literature.

Large-scale global “black-box” optimization: evolutionary metaheuristics for adaptive problem decomposition

In recent years, many real-world optimization problems have had to deal with growing dimensionality. Optimization problems with many hundreds or thousands of variables are called large-scale global optimization (LSGO) problems. Evolutionary algorithms (EAs) have proved their efficiency at solving many complex real-world optimization problems. However, their performance usually decreases when the dimensionality of the search space increases. Today, the best results and the majority of approaches are presented by so-called “decomposition methods’, that divide a LSGO problem into smaller parts by grouping its objective variables. EAs deals with subproblems using the cooperative coevolution framework. The finding of an appropriate decomposition is a part of the general search process. In the tutorial, we will discuss recent adaptive approaches and results in the field