- author = "Marcus Komann",
- title = "Manual Engineering and Evolution of Emergent Algorithms for Agents on Two-dimensional Grids",
- school = "Erlangen",
- year = "2010",
- type = "Doktor-Ingenieur",
- address = "Germany",
- month = "22 " # dec,
- keywords = "genetic algorithms, genetic programming, EHW, FPGA, agent, evolution, emergence, deterministic finite state machine, field programmable gate array, feature extraction, smart camera",
-
URL = "
http://www.opus.ub.uni-erlangen.de/opus/volltexte/2011/2315",
-
URL = "
http://www.opus.ub.uni-erlangen.de/opus/volltexte/2011/2315/pdf/MarcusKomannDissertation.pdf",
- size = "292 pages",
- abstract = "In this thesis, the problem of detecting the attributes of multiple objects in binary images in realtime is solved. It is a common problem in industrial machine vision. For the solution, the usage of emergent algorithms on a smart camera with a fine-grained massively-parallel processor is proposed. Combining both is promising since such processors can exploit the abilities of emergent algorithms. Therefore, so-called Marching Pixels are introduced. These are local agents that traverse the pixel grid of an image in a certain way in order to accumulate global data about the image objects. At first, manually engineered Marching Pixels algorithms for different object classes are presented and compared. Afterwards, their realisations in hardware are shown. These realizations are able to fulfil the real time requirements and are small enough for application in real industrial scenarios. They can further be used to execute other emergent algorithms that are based on 2-D grids. However, the absolute quality of the manually engineered algorithms is unknown. In the thesis, emergent agent algorithms for 2-D grids are thus also evolved. First, the effectiveness of evolution for finding good emergent agent algorithms is shown. Then, it is argued that improving agent abilities has to be done by cautiously balancing increased agent amounts and higher agent intelligence. The focus on the initial machine vision problem is thereby expanded to emergent agents on 2-D grids in general since many complex systems today comprise such entities and controlling their emergent phenomena is difficult but worthwhile.",
-
notes = "Some consideration of GP in later chapters.
Advisor: Fey, Dietmar (Prof. Dr.-Ing.)",
