Final Activity Report Summary - COMP2SYS (Computational intelligence methods for complex systems)
Computational Intelligence (CI) provides a set of techniques to understand, predict, optimise and control the behaviour of complex systems when the assumptions underlying conventional techniques are questioned. The research in the project focussed on the:
1. development and enhancement of existing CI techniques
2. application of CI techniques to a number of challenging problems that arose in a variety of disciplines ranging from robotics to optimisation and machine learning
3. study of fundamental issues related to the behaviour of complex systems.
The main research contributions of the doctoral students participating in the project could be summarised as follows.
In terms of optimisation, we developed new algorithmic techniques for tackling problems when some of the information was stochastic. New state-of-the-art algorithms were developed for various stochastic problems. Other work focussed on the parallelisation of metaheuristics, such as ant colony optimisation, and on the detailed study of the resulting performance improvements.
Regarding modelling of complex systems, the research focussed on biological and socio-economic systems. Its main contributions were models that offered novel insights into the emergence of topological characteristics observed in real biological networks and the study of the role of static and dynamic social ties in the emergence of cooperation.
With respect to analysis of complex and messy data, our research focussed on data analysis tasks arising in wireless sensor networks. In this area, prediction models were developed that allowed to strongly reduce energy consumption of the sensors and, in addition, to detect sensor failure. Other results concerned the development of distributed algorithms to compute specific statistical measures, i.e. principal components, which allowed for network load reduction.
Moreover, a main part of the research efforts focussed on the development of control strategies for several tasks arising in collective robotics, where a number of simple robots collaborated to perform tasks that were beyond their individual capabilities. Tasks that were successfully tackled included cooperative transport of heavy objects, development of efficient foraging strategies and morphology control for autonomous, self-assembling swarms of robots. Other results concerned the development of techniques for detecting failing members in a swarm of robots.
Finally, in terms of swarm intelligence, many of the developed control strategies and optimisation algorithms were inspired from social insect behaviour. Hence, these developments produced clear evidence that the relatively new research area of swarm intelligence, which was a novel approach to distributed control and distributed optimisation, was a promising subfield of computational intelligence research.
1. development and enhancement of existing CI techniques
2. application of CI techniques to a number of challenging problems that arose in a variety of disciplines ranging from robotics to optimisation and machine learning
3. study of fundamental issues related to the behaviour of complex systems.
The main research contributions of the doctoral students participating in the project could be summarised as follows.
In terms of optimisation, we developed new algorithmic techniques for tackling problems when some of the information was stochastic. New state-of-the-art algorithms were developed for various stochastic problems. Other work focussed on the parallelisation of metaheuristics, such as ant colony optimisation, and on the detailed study of the resulting performance improvements.
Regarding modelling of complex systems, the research focussed on biological and socio-economic systems. Its main contributions were models that offered novel insights into the emergence of topological characteristics observed in real biological networks and the study of the role of static and dynamic social ties in the emergence of cooperation.
With respect to analysis of complex and messy data, our research focussed on data analysis tasks arising in wireless sensor networks. In this area, prediction models were developed that allowed to strongly reduce energy consumption of the sensors and, in addition, to detect sensor failure. Other results concerned the development of distributed algorithms to compute specific statistical measures, i.e. principal components, which allowed for network load reduction.
Moreover, a main part of the research efforts focussed on the development of control strategies for several tasks arising in collective robotics, where a number of simple robots collaborated to perform tasks that were beyond their individual capabilities. Tasks that were successfully tackled included cooperative transport of heavy objects, development of efficient foraging strategies and morphology control for autonomous, self-assembling swarms of robots. Other results concerned the development of techniques for detecting failing members in a swarm of robots.
Finally, in terms of swarm intelligence, many of the developed control strategies and optimisation algorithms were inspired from social insect behaviour. Hence, these developments produced clear evidence that the relatively new research area of swarm intelligence, which was a novel approach to distributed control and distributed optimisation, was a promising subfield of computational intelligence research.