Title :
Downhill walk from the top of a hill by evolutionary programming
Author_Institution :
Dept. of Electr. & Electron. Eng., Anadolu Univ., Eskisehir, Turkey
Abstract :
When we search for an infinitely large number of solutions by evolutionary algorithms, it is helpful to learn the topology of the fitness landscape to know whether the solutions we obtained are representative samples of the whole solutions. Some solutions are easy to be approached and others are not in general. As a step to learn the whole geometry of fitness landscape, we exploit, in this paper, a downhill walk by evolutionary programming to reveal the shape of global peaks on the fitness landscape defined on weight space
Keywords :
evolutionary computation; learning (artificial intelligence); neural nets; downhill walk from the top of a hill; evolutionary programming; fitness landscape topology learning; global peak shape; search; weight space; Chemistry; Circuit topology; Computational geometry; Evolution (biology); Evolutionary computation; Genetic programming; Neural networks; Physics; Shape; Solid modeling;
Conference_Titel :
Evolutionary Computation, 1999. CEC 99. Proceedings of the 1999 Congress on
Conference_Location :
Washington, DC
Print_ISBN :
0-7803-5536-9
DOI :
10.1109/CEC.1999.782648
