Evolutionary Algorithms (EvA)

EvA is a program package for optimization with Evolutionary Algorithms, in particular Genetic Algorithms, Evolution Strategies and Simulated Annealing. The particular aim of developing EvA was a comparison between SIMD and MIMD parallel computer architectures. Orthogonally to the algorithms (GA, ES, SA) there exist massively parallel (MP) implementations (MPGA, CNGA, MPES) on SIMD computers and distributed implementations (VEGA, VEES, VESA) on MIMD computers with very high performance.

Screenshot of VEES with the user interface UIEA and a viewer for the Traveling Salesman Problem

EvA was presented with great success at the CeBIT '98 international exhibition in Hanover, Germany. Additionally it was shown at the fair "Wirtschaft trifft Wissenschaft" (industry meets science) in Stuttgart (Germany) with the Homag industry-application. EvA was also used for the practical exercises for the lecture "Genetic Algorithms and Evolution Strategies" which is offered since summer semester 1996.

When it was used in the projects with Homag and BMW EvA was extended in such a way that it was easy to apply external user interfaces, adopted for the particular optimization problems. In cooperation with the company "Homag Maschinenbau AG" (see section Homag), a restricted form of the optimization problem was dealt with by a diploma thesis by Hartmut Ott. This resulted in a contribution which was accepted by the 4th Symposium on "Soft Computing in Production and Material Industry" on March, 12, at the University of Göttingen. A talk entitled "Evolutionary Algorithms for Position Opimization of clamps for the fixation of wooden workpieces" was given at the symposium.

Current investment is on steady-state evolution strategies whereby the algorithm is not generation-based, but in one step of the algorithm, only one (or a few) individuals are generated. The individual is then immediately evaluated and integrated back into the population. Thus, this algorithm is especially suited for fitness functions with high computation costs, which is likely in real-world applications. An algorithm shall be developed which preserves the ability of fast self-adaptation, which is normally lost in steady-state algorithms.

The following diagram shows the structure of the EvA package:

The chair for Cognitive Systems is member of EvoNet.