|
| Project: | Nonstandard variation operators for evolutionary algorithms, subproject B3 of the collaborative research centre (SFB) 531
|
| Description: | The main research interest in this project is the design of problem-specific evolutionary algorithms. The aim is to systematize the design of evolutionary algorithms for problems with nonstandard representations. During the last years (1997-2002) , the project focused on nonstandard, problem-specific representations for structure optimization problems. Now, we pay special attention to the variation operators. We investigate miscellaneous methods to generate new search points (individuals).
Alternative models of evolutionary algorithms can provide new means for a systematic integration of domain-knowledge. The analysis of particle-swarm models and estimation of distribution algorithms is used as a starting-point. The long-term goal is to develop a suitable combination of evolutionary search heuristics and machine learning techniques in order to support the design of problem-specific evolutionary algorithms.
In addition, we investigate how evolutionary algorithms can be used in order to find robust solutions. Furthermore, we analyze the (self-)adaptation of strategy parameters by means of empirical studies.
|
|
| Boris Naujoks | Telephone: 0231 - 755 7705 | boris.naujoks  cs.uni-dortmund.de | |
|
| Project: | Genetic programming and neural networks, subproject B2 of the collaborative research centre (SFB) 531 |
| Description: | After examininig the combination (hybridization) of neural networks and genetic programming (GP) in the first years of the SFB, we recently concentrated more on GP itself. In particular, methodical developments in the area of linear GP have been dealt with.
|
|
|
| Project: | Bio-inspired learning and optimization methods for networks |
| Description: | Within the research project LEONET, natural computing methods are used for solving problems in complex and strongly interconnected systems. Concerning the application fields, the project focuses on road trafic, internet computing, and telecommunication networks. Concerning the natural computing methods used for tackling these applications, mainly fuzzy logic, neural networks and evolutionary algorithms are exploited.
|
|
|
| Project: | Learning of user preferences for agent systems, subproject KD of Bio-inspired learning and optimization methods for networks |
| Description: | Software agents are autonomous, active, and compact software components. They form an innovative design paradigm for the efficient realization of complex, heterogeneous software systems with a high rate of interaction. Software agents must be able to communicate with other agents and external components. At the ICD, learning multi-agent systems are used as basic components for a personal travel assistance (PTA) system and an electronic commerce system. The task is to realize system components that learn travel preferences of a user.
|
| Organizer: | Boris Naujoks | Telephone: 0231 - 755 7705 | boris.naujoks cs.uni-dortmund.de | |
| |
|
| Project: | Routing and dimensioning of telecommunica- tion networks,
subproject KP of Bio-inspired learning and optimization for networks |
| Description: | Within this part of the LEONET project, optimization methods based on evolutionary algorithms are developed for solving problems which are of paramount importance for the design and management of telecommunication systems.
The particular focus is on private telecommunication networks that realize the communication in large companies and public authorities.
|
|
|
| Project: | Design and dimensioning of power stations,
subproject AK of Bio-inspired learning and optimization for networks |
| Description: | This project, which has now been successfully completed, dealt with the conception of problem specific evolutionary algorithms (EA) for the coupled structure and parameter optimization of thermal power plant processes. Problem specific search methods have been developed and successfully integrated into the power plant design environment of the industrial partner, the Siemens AG.
|
|
|
| Project: | Theoretical investigations of evolutionary algorithms |
| Description: | It is the goal of this project, which is supported financially by the Deutsche Forschungsgemeinschaft (DFG), to further the theoretical understanding of the behavior of evolutionary algorithms and in particular of evolution strategies in real-valued search spaces. More specifically, it is determined how the performance of such strategies scales with parameters of the problem - such as the dimensionality of the search space - or of the strategy - such as the population size. By virtue of such scaling laws strategy variants can be compared, guidelines for tuning evolution strategies for maximum performance can be given, and insights and an understanding of the behavior of the strategies can be gained that goes beyond what can be learned from mere experimentation.
|
|
|
| Project: | Unsupervised learning in neural networks |
| Description: | A promising alternative for the widely-accepted supervised learning schemes for neural networks is the design of unsupervised learning procedures, which can learn to encode the systematic structure of their input data as network-internal representation. For building an internal representation of the external data world it is not necessary to give any desired responses to the network. While conventional data analysis usually requires the specification of modeling hypotheses the parameters of which have to be estimated, neural networks offer the chance to discover structure within data without the need for specifiying special models.
|
|
|
| Project: | Evolution of gait coordination with genetic programming |
| Description: | The goal of this project, which is part of the german priority program Autonomes Laufen, is to develop a method for automatically designing robot controllers for moving arbitrary robot hardware architectures. It should then be possible to explore the space of potential gait patterns of any robotic hardware with little or no need for a kinematic model of the robot's architecture.
|
|
|
| Project: | BinSys - Self-organization in binary-string systems |
| Description: | In the (now completed) project BinSys self-organization phenomena in artificial chemistries were studied. An artificial chemistry is roughly defined by a set of objects - the molecules - and a set of interaction rules. In our case the objects were mostly binary strings.The phenomena of evolution, information processing, and the problem of visualizing complex population dynamics are of special interest.
|
|
|
| Project: | Socionics |
| Description: | Socionics deals with the exploration and modeling of artificial societies. Socionics has one of its seeds in the field of artificial intelligence (AI). While the human brain is the origin of intelligence in classic AI, distributed artificial intelligence (DAI) assumes that the interaction of many acting individuals leads to the solution of a problem. Hence, the solution of a given problem is not the result of individual intelligence, but of social intelligence.
In our project we primarily investigate the dynamics of social systems and the modeling of complex agents. We follow two lines of research in parallel. First, we try to explain the dynamics of structures and system processes with learning and reflexive agents. Secondly, we develop an architecture to create complex agents for modeling social actors.
|
|
|
| Project: | Automatic optimization of selected chemical processes |
| Description: | In this project robust strategies for the automatic optimization of chemical process units are developed. By establishing such methods, shorter development intervals and a significantly higher productivity and quality concerning the intermediate products can be achieved. This strategy can contribute in an early stage to the optimization of the whole production system. For this purpose more or less general numerical and stochastic optimization methods are utilized, adopted and validated.
|
|
| |
|
| Project: | Organic evolution and evolutionary algorithms: Using further evolutionary principles, subject of the subproject B3 of the SFB 531 |
| Description: | In the last 40 years, nature-inspired heuristics imitating certain principles of modern synthetic theory of evolution have proved their capability in solving real world problems. In general, main effects of mutation, recombination and selection are used to mimic the process of micro evolution.
For the case of unimodal single objective problems, this level of abstraction was sufficient to establish these heuristics among conventional optimization methods. In case of multimodal problems or problems with constraints, however these population-based methods show their weaknesses. Stagnation of the evolution process is observable very often. If the optimization is extended to the multi objective case, the heuristics fail. Further - we call it here once careless - artificial improvements like density based selection or archiving techniques were added to the nature-inspired heuristics. These techniques are usually connected with the selection operator and become more and more complex and sophisticated. Why however should one deviate from the giving path of nature?
The primary aim of this project is to extract and transfer essential evolutionary principles in order to improve the design of evolutionary algorithms.
|
|
|
| Project: | DREAM - Distributed resource evolutionary algorithm machine |
| Description: | The project is funded by the European Union (EU) and aims at creating a framework for using available computing power of machines distributed on the Internet for applications from the domain of evolutionary computation.
Utilizing the DREAM framework, parallelized evolutionary algorithms are applied e.g. to scheduling problems and as machine learning methods.
|
| Organizer: | Mike Preuß | Telephone: 0231 - 755 7705 | mike.preuss cs.uni-dortmund.de | |
|
| Project: | Optimization of electronic circuit designs |
| Description: | The growing demand for high performance mobile electronic systems with enhanced battery lifetime requires the design of fast integrated circuits with very low power consumption. In cooperation with the Chair of Microelectronics at the University of Dortmund, the capabilities of different design choices were analyzed.
We investigated different (multi-)objective functions, e.g. power dissipation, signal delay, chip area, etc., that represent global, high-dimensional optimization problems. Multi-objective optimization evolutionary algorithms (MOEA) have been considered as promising approaches for these complex problems.
|
|
|
| Project: | Particle swarm optimization, subject of the subproject B3 of the SFB 531 |
| Description: | The term swarm intelligence is used to describe algorithms and distributed problem solvers that are inspired by the collective behavior of animal societies like insect colonies, schools of fish, or flocks of birds. Under this prism, particle swarm optimization (PSO) is a collective intelligence method for solving optimization problems.
In cooperation with Prof. Vrahatis (Artificial Intelligence Research Center UPAIRC of the University of Patras), we are developing methods to improve the search behavior of PSO. Since PSO and evolutionary algorithms (EA) are based on common principles, it might be interesting to develop a theoretical framework to enable the comparison and analysis of PSO and EA.
|
|
|
| Project: | Elevator group controller optimization |
| Description: | In cooperation with Fujitec Co., Ltd. (Japan) and NuTech Solutions GmbH (Germany), an elevator group control optimization problem is investigated. The elevator group control task is a real-time optimization problem of allocating elevator cars to passengers requesting service. Although it has been investigated for many decades, it is still an open research problem. The main difficulties lie in the stochastic nature of passenger arrivals, and in the combinatorial explosion of system states with the number of cars and floors.
|
|
|
| Project: | IKY-DA Projekt: Low Cost Optimisation with applications in turbomachinery design and chemical engineering |
| Description: | The IKY-DA project is a bilateral exchange project for PhD and diploma students between the ICD/CASA and the Dept. of Technical Turbomachinery of the NTUA Athens. The scientific goal of the project is the development of new fast optimisation tools for optimisation in turbomachinery and chemical reactor design, based on evolutionary algorithms and multivariate interpolation (metamodelling) algorithms that function as fast predictors for evaluation results.
|
|
|
| Project: | Adaptation and Dynamics in Evolutionary Algorithms
subproject A3 of the SFB 531 |
| Description: | This project focuses on a theoretical analysis of evolutionary
algorithms (especially evolution strategies) for optimization problems
on continuous search spaces.
To enable the algorithm to approach the optimum, the strategy parameters
of the algorithm have to be adapted accordingly. One research topic
is therefore an investigation and comparison of different adaptation
mechanisms.
Another aim is an analysis of the behavior of
evolution strategies under noisy fitness evalutations since
noise is a common problem in real-world optimization tasks.
Since dynamic optimization is frequently cited as a main application area
for evolutionary algorithms, the tracking behavior of evolution strategies
is also investgated.
|
|
|
| Project: | Data-based generation and optimization of fuzzy-models using the Fuzzy-ROSA method, subproject T1 of the transfer unit (TFB) 37 |
| Description: | For complex technical systems, physical or knowledge-based modeling is often
very time-consuming or even fails. In such cases a data-based modeling approach
using available system data may be successful. In particular, fuzzy models have
proven to be suitable for this, as they can be generated from available data
and can be improved by adding rules generated from expert knowledge. Fuzzy
models distinguish themself by the interpretability of the resulting model,
since they consist of qualitative if-then rules, which corresponds to the way
human knowledge is often presented. The Fuzzy-ROSA method generates only rules
which have passed a statistical test based on the underlying data. Consequently,
each accepted rule is reasonable in sense of the applied test. In view of a
growing demand for modeling complex systems different methods applying to the
Fuzzy-ROSA method have been developed in the subproject B1 of the collaborative
research center (SFB) 531. Purpose of the subproject T1 is the systematic
exploration and prototypical implementation of these methods.
|
|
|
| Project: | APOMAT Netzwerk: Optimierung in der Materialforschung |
| Description: | APOMAT 526 is a thematic network that consists of
several small joint projects that is evaluated and partially financed
by the European research comission within the COST ACTION framework.
The main objective of APOMAT 526 is to develop and to apply numerical optimization methodologies
for automatic materials process design, based on quantified product qualities,
relating to process targets and constraints, including economic aspects.
Collaborative work is based on evaluated projects and constitutes to a
high degree on interdisciplinary cooperation between European materials engineers and
optimization experts of high reputation.
|
|
Home
ls11.cs.uni-dortmund.de>
