New Generation Computing, 10 (1992) 423-427 OHMSHA, LTD, and Springer-Verlag News (~) OHMSHA, LTD. 1992 New Topics in Genetic Algorithm Research A. KONAGAYA C & C Systems Research Labs., NEC Corporation t-1, Miyamaeku, Kanagawa 216 Japan E-mail.konagaya@esl. el. nee. co. jp This paper reports on some new topics in Genetic Algorithms (GA) mainly focusing on the 4th International Conference of Genetic Algorithms s> held at San Diego on August 1991. Genetic algorithms 1>-4> or evolution strategiesm are probabilistic algortithms based on the biological evolution process. Unfortunately, most computer scientists have not paid much attention to genetic algorithms. However, ICGA-9 l seems to have had a great impact to the computer science community, at least in Japan. Some people are now wondering if genetic algorithms might become one of the popular AI techmologies like neuro computing and fuzzy logic. Of course, it is too early to judge the effectiveness of genetic algirithms at this time. However, it should be noted that recent GA activities emphasis their practical aspects rather than modeling nature. The otganization of this paper is the following. First, section 2 introduces a simple genetic algorithm for those who are not familiar with genetic algorithms. Then, section 3 introduces GA theories about deceptive functions which are difficult to optimize with GAs. Section 4 introduces the extended GAmodels which are proposed to optimize deceptive functions. Section 5 introduces problem specific GA models which are essential to solve real world problems by GAs. Section 6 summarizes parallel GA systems which actually run on parallel computers. w Simple Genetic Algorithm The simple genetic algorithm (SGA) is a basic model of a genetic algorithm 1>. SGA simulates the survival of the fittest in a population of individuals which represent points in a search apace. The individuals are often represented by binary strings. A function, often called a fitness function, gives values to the binary strings. The aim of a gentic algorithm is to find a global optimun of the fitness function when given an initial population of individuals by applying genetic operators in each generation, a period in which the individuals can survie. The genetic operators consist of the following operators: crossover, mutation and selection. 1.1 Crossover The crossover operator produces two descendants by exchanging part of two individuals, This operator aims to make a better individual by replacing a part of an individual with a better part of another individual. For example, crossover of the strings "000110" and "110111" at the third postion produces the strings "000111" and "110110". The candidates of the crossover operation and the crossover position are