Cultural Transmission of Information in Genetic Programming
نویسندگان
چکیده
This paper shows how the performance of a genetic programming system can be improved through the addition of mechanisms for non-genetic transmission of information between individuals (culture). Teller has previously shown how genetic programming systems can be enhanced through the addition of memory mechanisms for individual programs [Teller 1994]; in this paper we show how Teller’s memory mechanism can be changed to allow for communication between individuals within and across generations. We show the effects of indexed memory and culture on the performance of a genetic programming system on a symbolic regression problem, on Koza’s Lawnmower problem, and on Wumpus world agent problems. We show that culture can reduce the computational effort required to solve all of these problems. We conclude with a discussion of possible improvements. 1 Culture and Evolution In most evolutionary computation systems, individuals are assessed for fitness independently. Each individual is placed into an initialized environment, a fitness test is performed, and the environment is then re-initialized for the next individual. By contrast, in biological populations (and in many Artificial Life systems; see, e.g., [Brooks and Maes 1994]) each individual acts in a shared environment that may bear the marks of contemporaries and of predecessors. Some animals use these marks to their advantage, and some intentionally modify the environment to communicate with others. Modifications to the environment may be ephemeral (e.g., calls, spoken language) or long-lasting (e.g., nests, written language). Through these mechanisms fitness-enhancing information may be transmitted to others, including offspring, by non-genetic means. We use the term “culture” to refer to any information transmitted between individuals by non-genetic means. Our 0To appear in the Genetic Programming 96 (GP96) conference proceedings, Stanford, July 1996. definition is similar to that of Bonner, who describes his use of the term as follows: By culture I mean the transfer of information by behavioral means, most particularly by the process of teaching and learning. It is used in a sense that contrasts with the transmission of genetic information passed by the direct inheritance of genes from one generation to the next. [Bonner 1980, p. 9] Culture, in this sense, is employed by many animals other than humans. Higher primates provide many examples, but one can also argue that much simpler animals, even insects, make significant use of culture [Bonner 1980]. One can view the evolution of a culture-using species as a complex interaction between two processes, one genetic and one cultural. Dawkins coined the term “meme” to serve the function for cultural transmission that “gene” serves for genetic transmission; that is, a meme is a unit of information transmitted by behavioral means [Dawkins 1976]. The interactions between cultural and genetic transmission, and their combined effects on fitness and adaptation, have been the subject of much discussion in evolutionary biology. For our purposes we may simply note that the difference in mode of transmission leads to a significant difference in the speed with which a new piece of information may spread through a population. Again, quoting Bonner: If, for instance, a favorable genetic mutation appears in one individual, even with reasonable positive selection pressures, it will take many life cycles and perhaps hundreds of years for the new gene to be present in an appreciable number of individuals in the population. It is only in some lethal circumstances that the genetic structure of a population can change rapidly. ... On the other hand a cultural change can be exceedingly rapid. A new fad or dress style may take over a whole nation in a matter of days or weeks. [p. 18] The difference in speed of transmission leads to a complementary difference in stability; the gene pool is less vulnerable to sudden dangerous fluctuations than is the meme pool. As a result one can expect that a combination of the two transmission modes will be most beneficial for the evolution of certain complex systems. Several researchers have previously experimented with cultural elements in evolutionary computation systems. For example, Bankes has explored “meme-based” adaptive systems in multiplayer games, although the bulk of this work is concerned with the evolution of memes within individuals [Bankes 1995]. Reynolds has developed a framework of dual inheritance “cultural algorithms” in which a “belief space,” containing generalizations of the representations of individuals, helps to guide the evolution of a population [Reynolds 1994]. Others have explored related ideas in a range of evolutionary computation contexts (e.g., [Hutchins and Hazlehurst 1993]). In contrast to previous research, this paper describes a cultural mechanism that is a straightforward extension of the memory mechanisms used by individuals. The same mechanisms that are used by individuals to build and maintain state information are extended to allow for the communication of information between individuals within and across generations. The virtues of this scheme include simplicity, a better match to biological notions of culture, and a demonstrable positive impact on the speed of evolution for several test problems. The remainder of this paper is organized as follows: after a brief digression on performance in genetic programming we describe a simple technique for adding cultural transmission to genetic programming systems. We then show how the additionof culture can reduce the computational effort required to produce solutions for symbolic regression, Lawnmower, and Wumpus world problems. We conclude with a discussion of possible improvements to the technique. 2 Genetic Programming and
منابع مشابه
Network Planning Using Iterative Improvement Methods and Heuristic Techniques
The problem of minimum-cost expansion of power transmission network is formulated as a genetic algorithm with the cost of new lines and security constraints and Kirchhoff’s Law at each bus bar included. A genetic algorithm (GA) is a search or optimization algorithm based on the mechanics of natural selection and genetics. An applied example is presented. The results from a set of tests carried ...
متن کاملA Novel Method for Designing and Optimization of Networks
In this paper, system planning network is formulated with mixed-integer programming. Two meta-heuristic procedures are considered for this problem. The cost function of this problem consists of the capital investment cost in discrete form, the cost of transmission losses and the power generation costs. The DC load flow equations for the network are embedded in the constraints of the mathematica...
متن کاملAn Efficient Predictive Model for Probability of Genetic Diseases Transmission Using a Combined Model
In this article, a new combined approach of a decision tree and clustering is presented to predict the transmission of genetic diseases. In this article, the performance of these algorithms is compared for more accurate prediction of disease transmission under the same condition and based on a series of measures like the positive predictive value, negative predictive value, accuracy, sensitivit...
متن کاملUsing Genetic Algorithm in Solving Stochastic Programming for Multi-Objective Portfolio Selection in Tehran Stock Exchange
Investor decision making has always been affected by two factors: risk and returns. Considering risk, the investor expects an acceptable return on the investment decision horizon. Accordingly, defining goals and constraints for each investor can have unique prioritization. This paper develops several approaches to multi criteria portfolio optimization. The maximization of stock returns, the pow...
متن کاملRanking the Efficiency of Job Performance of Employees and Managers through Cultural Intelligence based on the DEA Linear Programming Model
Purpose: The job performance of employees and managers plays an important role in promoting the organization's performance, and it seems that cultural intelligence can affect it. As a result, the aim of this study was to ranking the efficiency of job performance of employees and managers through cultural intelligence based on the DEA linear programming model. Methodology: This study in terms o...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 1996