Gruau, F. and Whitley, D. (1993) Adding learning to the cellular development of neural networks: Evolution and the Baldwin effect. Evolutionary Computation 1(3):213-233.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....search are investigated. 2 Attempts in Combining Local and Evolutionary Search In the belief that better results can be achieved by combining local search and evolutionary search, various attempts have been made to adopt this synergetic approach to construct and train neural networks. Some [8,25,39,49] achieved good results while others [41,57] found that the resulting hybrid algorithms are not efficient. These attempts differ in how local search is applied, and the differences are summarized in this section. 2.1 Nature of Local Search Local search aims at searching for better solutions in ....

....x 3 is the worst. The ith bit of offspring x 4 (denoted as x 4i ) is set to x 1i if x 1i = x 2i ; otherwise x 4i is set to the negation of x 3i . Although the local search method is very simple, the inclusion of this operator is found to be able to improve the evolution process. Gruau and Whitley [25] proposed a local search method and compared different approaches to combining local search and evolutionary search. In their hybrid algorithms, a boolean neural network is represented by a grammar tree (instead of a string of floating point numbers or a binary string) that specifies the number of ....

[Article contains additional citation context not shown here]

F. Gruau and D. Whitley. Adding learning to the cellular development of neural networks: Evolution and the Baldwin effect. Evolutionary Computation, 1(3):213--233, 1993.

....a complete copy of the genome . To achieve the degree of exibility required when any cell can have any function, recon gurable technology may be said to be a requirement. This is especially the case since the cellular structure varies as a function of the application [11] Cellular encoding [12] is a well de ned formal approach based on biological development. This approach may be used to build complex systems by encoding the cell types, timing of cell division and changes in links involving the cell [13] This methodology is a variation of genetic programming [14] The work of ....

F. Gruau and D. Whitley. Adding learning to the cellular development of neural networks: Evolution and the baldwin eect. Journal of Evolutionary Computation, 1993.

....to study a variety of effects in GAs such as exploration vs. exploitation, and their effects on residual learning [5] Neural networks have also been use in the investigation of the Baldwin effect, and there have been studies incorporating Baldwin style effects in the evolution of neural networks [6]. Such studies demonstrate that the criticisms of the learning framework used by Hinton and Nowlan do not fundamentally undermine the effects shown. In short, the Baldwin effect is portable to more complex learning mechanisms. The Baldwin effect has been replicated in an artificial life ....

F. Gruau and D. Whitley, "Adding learning to the cellular devel- opment of neural networks: Evolution and the baldwin effect," Evolutionary Computation, vol. 1, no. 3, pp. 213-233, 1993.

....learning [20] to name a few. However, understanding the original focus of Hinton and Nowlan s simulation the Baldwin e#ect is still a topic of discussion. It has been studied with a classic population genetics approach [1, 6] and with boolean neural networks based on grammar tree encodings [8]. Studies concerning the influence of evolutionary operators on genetic assimilation [11, 12] as well as comparisons between Lamarckian and Baldwinian forms of evolution [22] have also been undertaken. But the fundamental issue of why genetic assimilation of learned behaviour occurs at all within ....

Gruau, F. and Whitley, L. D. (1993). Adding learning to the cellular development of neural networks: Evolution and the Baldwin e#ect. Evolutionary Computation, 1(3):213--233.

....where chaos can introduce itself. This approach is now well known, and much work has been done (see [3] for a resume of research in neurogenesis) The morphogenesis process may make use of production rules at the neuron level, such as Gruau who uses a grammar tree as rewriting rules for neurons ([4]) It may be implemented on a cellular automata ( 5] Finally, it can make use of production rules for cellular elements such as proteins ( 6] 7] 8] It is with this approach that we address the evolution of neural network. II. Artificial Neurogenesis The method we developed is inspired by ....

F. Gruau and D. Whitley, "Adding learning to the cellular development of neural networks: Evolution and the baldwin effect", Evolutionary Computation, vol. 1, no. 3, pp. 213-- 234, 1993.

....the development of the control system of the various motor subsystems (such as legs) of an animat. Furthermore, phenotypes can be exposed to learning in an environment for fitness determination thus allowing for the study of the interaction between learning and evolution [Hinton and Nowlan, 1987; Gruau and Whitley, 1993] 127 Summary A performance comparison of ADF and module acquisition (MA) as well as other variations of the two methods, is presented in [Kinnear Jr. 1994] ADF consistently shows better performance. These is attributed to the repeated use of calls to automatically defined functions and to ....

Frederic Gruau and Darrel Whitley, "Adding Learning to the Cellular development of Neural Networks: Evolution and the Baldwin Effect," Evolutionary Computation, 1(3):213--233, 1993.

....(see [16] that is, a multiple agent genetic programming implementation in which agents are evaluated on both individual performances and joint operation on a shared object. Each agent generates instruction sets rather than musical output. Borrowing from Gruau s work on cellular encoding (see [11]) we generate musical output with the help of a seeded cantus firmus; this output is a direct manifestation of the agents joint operation. Section 4.4 further details the interaction between agents. 2. Architecture 2.1 Individual and Subpopulation Overview Each agent consists of one ....

....functions. Function sets for the ADFs include , and variable specific variable setting functions. As with all other agents, after initial testing and development of sufficient function and terminal sets, the Pange Lingua Gloriosi plainchant became the only fitness case. 7 4. 4 Embryo Gruau [11] has described a method in which genetic programming specifies the graphconstructing operations necessary in growing an embryonic neuron into a full neural network. Koza et al. 17] broadened Gruau s concept so that genetic programming specifies graph constructing operations necessary in growing ....

Gruau, F., D. Whitley, 1993. "Adding learning to the cellular development of neural networks: Evolution and the Baldwin Effect," Journal of Evolutionary Computation, 1(3), pp. 213-- 233.

....capable, in principle, of optimizing any classification structure or set of structures. In fact, people have tried optimizing most traditional machine learning structures as well as some nontraditiona l structures using GAs. These structures have ranged from neural network weights and topologies (Gruau Whitley, 1993; Whitley et al. 1990, 1991, 1993; Whitley Schaffer, 1992) to LISP programs (Koza, 1992) to regions of the instance space similar to decision trees induced by a splitting algorithm (Rendell, 1983, 1985; Sikora Shaw, 1994) to expertsystem rules (Montana, 1990) to weights for a game s ....

Gruau, F., and D. Whitley. 1993. Adding learning to the cellular development of neural networks: Evolution and the Baldwin effect. Evolutionary Computation 1(3):213233.

....evolutionary algorithm. Recipe representations In this approach it is not a complete network that is coded into a chromosome, but a description of the growth process of a network architecture. This description usually is in the form of some kind of graph grammar describing the states of a network [6, 17, 18, 25]. The philosophy behind these recipe representations using (graph) rewriting systems is the scalability of the coding, which can not be achieved when using blueprint methods. When using blueprint codings, problems needing large neural network architectures will need very large chromosomes. The ....

....[28] When using this formalism in a genetic algorithm, each member of the population contains a set of production rules from which the architecture of a feedforward network grows. Section 1.4 will describe the G2L system in detail. Gruau proposed a similar approach using cellular encoding [17, 18]. In his method the tree representation of genetic programming is used to store grammar trees that contain instructions which describe the architecture as well as the (discrete) weights of the network. The consequence of coding the weights is that all weights conform to the same layout when ....

[Article contains additional citation context not shown here]

F. Gruau and D. Whitley. Adding learning to the cellular development of neural networks: Evolution and the Baldwin effect. Evolutionary Computation, 1:213--233, 1993.

....been trained using back propagation. However, in this case, the computation cost can be so high as to make genetic algorithms impractical except for optimizing small topologies. Many researchers take developmental and evolutionary approach for designing neural networks. Boers ( boers93] Gruaru([gruaru93]) et al. Proposed the design method of neural networks based on L system and genetic algorithms. De Garis( garis94] has studied cellular automata and genetic programming based artificial brain. Sugisaka( sugi98] also has developed artificial brain. To develop more big and complex system, we ....

Gruaru, F. Whitley, D. "Adding Learning to the Cellular Development of Neural Networks: Evolution and the Baldwin Effect," Evolutionary Computation, vol. 1-3, pp. 213-233. 1993

....being embedded in the cellular GA. We have demonstrated that good performance can be achieved by embedding it in evolutionary search, suggesting that local search methods need not be sophisticated in order to obtain the benefit of combining evolutionary search and local search. Other researchers [11], 21] also demonstrate that the combination of simple local search and evolutionary search can yield better performance as compared to evolutionary search alone. However, the local search method adopted in [11] has limitations, as it is only applicable to simple Boolean networks with binary ....

....the benefit of combining evolutionary search and local search. Other researchers [11] 21] also demonstrate that the combination of simple local search and evolutionary search can yield better performance as compared to evolutionary search alone. However, the local search method adopted in [11] has limitations, as it is only applicable to simple Boolean networks with binary rather than floating point weights. As the parity problem used in [11] is simple enough to be solved by local search alone, good results were expected when it was combined with evolutionary search. In contrast, we ....

[Article contains additional citation context not shown here]

F. Gruau and D. Whitley. Adding learning to the cellular development of neural networks: evolution and the Baldwin effect. Evolutionary Computation, 1(3):213--233, 1993.

....of acquired or learned characteristics that are well adapted to the environment. The improved individual is placed back into the population and allowed to compete for reproductive opportunities. However, Lamarckian learning inhibits the schema processing capabilities of genetic algorithms [8,21,22]. Changing the genetic information in the chromosomes results in a loss of inherited schema, altering the statistical information about hyperplane partitions implicitly contained in the population. While Lamarckian learning may disrupt the schema processing of a genetic algorithm, Baldwinian ....

F. Gruau and D. Whitley. Adding learning to the cellular development of neural networks. Evolutionary Computation, 1:213--233, 1993.

....to a hybrid optimization algorithm in which the efforts of local search (learning) and cellular GAs (stochastic optimization) are combined. In this paper, we explore the effects of two approaches of embedding learning into GAs, namely the Lamarckian and the Baldwinian 1 learning mechanisms [1] [3], 4] 7] In the next section, we describe the use of cellular GAs to evolve RNNs. Section III explains how we incorporated learning into the cellular GA. A long term dependency problem is described in section IV. In section V, we compare and K. W. C. Ku and M. W. Mak are with the Department of ....

....or the truncated BPTT(h;h 0 ) algorithm for one epoch, where an epoch is a complete presentation of all training patterns. Two different approaches of embedding learning into the cellular GA have been investigated in this study. They are the Lamarckian and the Baldwinian mechanisms [1] [3], 4] 7] A. The Lamarckian mechanism In the Lamarckian mechanism, the genotypes are modified by learning in order to improve the fitness of the chromosomes. The improvement is therefore automatically passed to these chromosomes. The idea is that the learnt behavior can directly change ....

F. Gruau and D. Whitley. Adding learning to the cellular development of neural networks: evolution and the Baldwin effect. Evolutionary Computation, 1(3):213--233, 1993.

....necessary number of cells. Moreover it may also be of use to generate a robust system, i.e. parts that have failed can grow again. Several developmental systems have been developed. The most notable works are by Kitano [5] and Fleischer [1] which concentrated on bio realism, and by Gruau et al. [3] and Kitano [6] which were grammar based developmental systems. The latter are closer to the developemental process we wish to adopt, but none of these works addressed the questions of efficiency, robustness and self programming, which are our main concerns. Thus our system will be developed along ....

F. Gruau and D. Whitley. Adding learning to the cellular development of neural networks: Evolution and the baldwin effect. Evolutionary Computation, 1(3):213--233, 1993.

....is embedded in cellular GAs to optimize the weights of RNNs. Finally, November 13, 1998 DRAFT 5 we conclude in Section VIII. II. Background Various attempts have been made to combine GAs and learning for the optimization of the weights and or topologies of neural networks. Some researchers [9] [17], 23] 30] achieved good results while others [24] 35] found that learning could not help much. Their experiments differ in how learning is applied. Some researchers [23] used GAs to find possible regions containing the global optimum, then used learning as a final fine tuning operator. Good ....

....be achieved by applying learning to chromosomes with better fitness. We believe that learning should be applied equally and that allowing poorly performed chromosomes to learn could also improve the evolution of the whole population. Therefore, we have adopted the approach similar to that of [9] [17] where learning was applied to fine tune every chromosome generated in each cycle of GAs. In our experiments, we have also investigated the effect of varying the learning frequency on the evolution process, as in [18] Usually, learning methods depend very much on the chromosomal representation. ....

[Article contains additional citation context not shown here]

F. Gruau and D. Whitley. Adding learning to the cellular development of neural networks: evolution and the Baldwin effect. Evolutionary Computation, 1(3):213--233, 1993.

....In the experiment of the Baldwin effect, the results of learning of individuals do not change their chromosomes, but only affect the selection after fitness evaluation. However, it is reported that incorporation of learning results into chromosomes also enhances the performance of GAs (see [9, 10] for example) This is known to be the Lamarckian inheritance. We also investigate the effect of Lamarckian inheritance on the evolution of associative memory, though a different implementation is required where the chromosomes are made up of components of the weight matrix instead [11] and we ....

Gruau, F., Whitley, D. (1993) `Adding Learning to the Cellular Development of Neural Networks: Evolution and the Baldwin Effect' Evolutionary Computation 1(3), pp213-233.

....the fitness landscape, but the nature of this form of evolution is still Darwinian. This effect is known as the Baldwin effect, after Baldwin who first proposed the idea a hundred years ago that learned behaviour of organisms could influence evolution [2] Figure 4 8, taken from Gruau and Whitley [26], illustrates how local optimization can alter the fitness landscape. N steps of local optimization deepens the basin of attraction, therefore making the landscape flatter around the local optima (minima for this example) When each individual always learns until it fully converges to a local ....

F. Gruau and D. Whitley; `Adding learning to the cellular development of neural network: evolution and the Baldwin effect'. In: Evolutionary Computation 1, 213-233, 1993.

....however, are not well understood yet. Here we tried to improve the memory capacity by using a genetic algorithm. Since the late 1980 s there have been a lot of papers dealing with neural networks by genetic algorithms (see [8] Most of them, however, were for layered neural networks (see [9, 10] for example) and only a few were for mutually connected neural networks [11] In this paper, we applied a simple genetic algorithm to mutually connected neural networks, and we have found that it can improve the capacity of associative memory. 3 8916 5 Takayama, Ikoma, Nara 630 01 Japan We ....

F. Gruau, and D. Whitley(1993). Adding Learning to the Cellular Development of Neural Networks Evolution and Baldwin Effect. Evolutionary Computation:213-233.

....can potentially offer significant advantages over single strategy systems. Since the type of input and acquired knowledge are more flexible, such hybrid systems can be applied to a wider range of problems. Examples of such integration include combinations of genetic algorithms and neural networks [Gruau and Whitley, 1993] and genetic algorithms and rule based systems [Bala et al., 1994] Vafaie and De Jong, 1994] The integration of genetic algorithms and inductive decision tree learning for optimal feature selection and pattern classification is a novel application of such an approach and is the topic of this ....

F. Gruau and D. Whitley, Adding Learning to the Cellular Development of Neural Networks: Evolution and the Baldwin Effect, Evolutionary Computation, Vol.1, No.3, pp. 213-234, 1993.

....attempts to bypass it, rather harm the efficiency of the GANN system [Hancock, 1992] Baldwin Effect GANN systems optimize neural networks in two ways: evolution by the genetic algorithm and learning by back propagation. There are a couple of possibilities to combine both elements. In [Gruau, 1993] the two main varieties are defined as Lamarckian learning and Baldwin learning. Yet, there are also more sophisticated approaches to combine GA and backpropagation, such as the ones proposed in [McInerrney, 1993] In the first case, the weights that are improved by the evaluation process are ....

....improved by the evaluation process are coded back into the chromosome. This undermines the ability of the genetic algorithm to perform hyperplane sampling, since the codes are altered during evaluation. Still, in the case of cellular encoding (see section 2. 3) it improves the GANN system the most [Gruau, 1993]. The second case is more close to natural Darwian evolution and profits from the Baldwin effect, which was first described in the 19th century. It impact on genetic algorithm was first suggested in [Hinton, 1987] though this report was criticized by Belew [Belew, 1989] In Baldwin learning, ....

[Article contains additional citation context not shown here]

: Frederic Gruau and Darrel Whitley: "Adding Learning to the Cellular Development of Neural Networks: Evolution and the Baldwin Effect", in: Evolutionary Computation, No. 1, Vol. 3, pp. 213-233, MIT Press.

....Selection Systems that employ several strategies have been shown to offer significant advantages over single strategy systems in other contexts. Examples of such systems include ensembles of neural networks and decision trees (Gutta and Wechsler, 1995) genetic algorithms and neural networks (Gruau and Whitley, 1993), and genetic algorithms and rule based systems (Bala et al., 1994; Bala and Wechsler, 1995; Vafaie and De Jong, 1994) The integration of genetic algorithms and decision tree learning described in this paper is designed to allow evolution and learning to work synergistically in a variety of ways. ....

Gruau F., and D. Whitley. (1993). Adding Learning to the Cellular Development of Neural Networks: Evolution and the Baldwin Effect, Evolutionary Computation, Vol.1, No.3, pp. 213-234.

....and exploiting directional trends in the fitness landscape, well advanced in the context of ESs, and or the corresponding operators, is another important avenue for research. Combinations of genetic search and local optimization resulting in either Lamarckian or developmental Baldwin learning (Gruau and Whitley, 1993) may also provide a means of addressing the difficulties imposed by ridge shaped landscapes. The question of which fitness assignment method is better remains largely open, although Pareto based methods seem more promising for their lack of sensitivity to the possible concavity of the trade off ....

Gruau, F. and Whitley, D. (1993). Adding learning to the cellular development of neural networks: Evolution and the baldwin effect. Evolutionary Computation, 1(3):213--233.

....ability. There are, however, theoretical objections to the use of a local search heuristic. An important one is that changing the genetic material in the population in a nonevolutionary manner will affect the schema represented in the population and undermine the GA. Gruau and Whitley [35] comment: Changing the coding of an offspring s bit string alters the statistical information about hyperplane subpartitions that is implicitly contained in the population. Theoretically, applying local optimization to improve each offspring undermines the genetic algorithm s ability to search via ....

F. Gruau and D. Whitley. Adding Learning to the Cellular Development of Neural Networks: Evolution and the Baldwin Effect. Evolutionary Computation, 1(3):213--233, 1993.

....of acquired or learned characteristics that are well adapted to the environment. The improved individual is placed back into the population and allowed to compete for reproductive opportunities. However, Lamarckian learning inhibits the schema processing capabilities of genetic algorithms [13, 40, 41]. Changing the genetic information in the chromosomes results in a loss of inherited schema, altering the statistical information about hyper plane partitions implicitly contained in the population. While Lamarckian learning may disrupt the schema processing of a genetic algorithm, Baldwinian ....

F. Gruau and D. Whitley. Adding learning to the cellular development of neural networks. Evolutionary Computation, 1:213--233, 1993.

....shifted toward the exploration of neural network architectures (see e.g. 11] 12] However, to represent architectures effectively, more sophisticated Genetic Algorithms had to be considered. For example, grammatical representations of the architecture of neural networks were proposed [10] [13]. Recently more biological approach was reported [14] All of these researches were for layered neural networks, and only a few were for Hopfield networks [15] 16] probably because of the fact that Hopfield network has a fixed architecture. However, sparse connectivity has a possibility to ....

F. Gruau, and D. Whitley (1993). Adding Learning to the Cellular Development of Neural Networks: Evolution and Baldwin Effect. Evolutionary Computation. 1(3) 213-233.

....of feed forward networks that are trained by supervised learning procedures and that could be used to control the behavior of an animat. Such a procedure has been used, for instance, by Pomerleau [28] to control the NAVLAB, i.e. the autonomous vehicle of CMU. 3. 2 Gruau The work of Gruau [29, 30] also encodes a rewriting grammar in a chromosome. However, this encoding scheme called cellular encoding rewrites neurons instead of symbols. In its simplest version, it is used to develop feed forward networks of Boolean neurons with integer thresholds and 1 or 1 connections, but more ....

....network evolved in Gruau s approach that controls tripod gait. AEP and PEP are sensory input units detecting anterior extreme positions and posterior extreme positions of the legs. PS and RS are motor output units controlling the power stroke and the return stroke (after [32] Gruau and Whitley [30] have added a variety of Hebbian learning to cellular development and evolution. In particular, following Hinton and Nowlan [34] and Belew [35] they have compared results obtained with fitness evaluations depending on a developed neural network alone to results obtained with fitness evaluation ....

F. Gruau and D. Withley. Adding learning to the cellular development of neural networks: Evolution and the Balwin effect. Evolutionary Computation, 1(3):213--233, 1993.

.... When learning is part of the fitness evaluation when searching for a good set of weights for a given architecture, a significant speed up and final quality of solution can be achieved [2, 15] Also when using evolutionary computation to optimize architectures, learning can increase performance [3,12], but sofar these attempts have been restricted to learning weights. With the algorithm presented in this paper, it will be possible to optimize modular artificial neural network architectures, implementing the Baldwin effect not just by learning weights, but by adapting the modular structure ....

F. Gruau and D. Whitley; `Adding learning to the cellular development of neural networks: evolution and the Baldwin effect'. In: Evolutionary Computation, 1, 213--233, 1993.

....into an unmanageable mess. For ERL methods to handle such problems, efficient mechanisms for adaptive behavior based on immediate rewards need to be developed. Several researchers have investigated the use of local learning after each action in an evolutionary algorithm (Grefenstette 1991; Gruau and Whitley 1993; Littman 1995; Nolfi and Parisi 1995) Such approaches have been termed Lamarckian if the local revisions are written back to the genetic chromosomes or Baldwinian if the revisions are not persevered. Local learning affords the evolutionary algorithm a quicker response to environmental shifts, ....

Gruau, F., and Whitley, D. (1993). Adding learning to the cellular development of neural networks.

....of acquired or learned characteristics that are well adapted to the environment. The improved individual is placed back into the population and allowed to compete for reproductive opportunities. However, Lamarckian learning inhibits the schema processing capabilities of genetic algorithms [9,31]. Changing the genetic information in the chromosomes results in a loss of inherited schema, altering the statistical information about hyperplane partitions implicitly contained in the population. Although schema processing may not be an issue given the higher order 5 cardinality ....

F. Gruau and D. Whitley. Adding learning to the cellular development of neural networks. Evolutionary Computation, 1:213--233, 1993.

....This helps the genetic search to converge towards genotypes that contain only the good rule. This feature, the Baldwin effect, has been more widely studied in genetic function optimisation (Whitley et al. 1994) and in the evolution of neural networks (Hinton and Nowlan, 1987; Belew et al. 1990; Whitley and Gruau, 1993) than in PCSs. 2.1.4 Cooperation among Rules The goal of any CS is to learn a KB, i.e. a set of rules, that controls the behavior of an agent so that it maximises the environmental reinforcement. This means discovering and preserving several alternative rules, which fire in distinct situations ....

Whitley, D. and Gruau, F. (1993). Adding learning to the cellular development of neural networks. Evolution and the Baldwin effect. Evolutionary Computation, 1.

....influence the neural development because a neuron is allowed to grow its branching axon only if the neuron s activation variability which depends upon the variability of the environmental stimulations to the network exceeds a genetically specified threshold. 5 Gruau The work of Gruau [GRUA92, GRUA93], like that of Boers and Kuiper, encodes a rewriting grammar in a chromosome. However, this encoding scheme called cellular encoding rewrites neurons instead of characters. In its simplest version, it is used to develop feedforward networks of Boolean neurons with integer thresholds and 1 ....

....capable of generating a sub network that solves a sub problem, then, of producing and combining copies of this sub network to build a higher level network that solves the problem. The genome splicing technique advocated by Koza [KOZA94] seems especially useful for such a purpose. Gruau and Whitley [GRUA93] have added a variety of Hebbian learning to cellular development and evolution. In particular, following Hinton and Nowlan [HINT87] and Belew [BELE89] they have compared results obtained with fitness evaluations depending on a developed neural network alone to results obtained with fitness ....

F. Gruau and D. Whitley. Adding learning to the cellular development of neural networks: Evolution and the Baldwin effect. Evolutionary Computation. 1, 3:213234. 1993.

....is to allow learning (i.e. local optimization) to change the fitness of an individual without altering the individual s genetic code. Local search, or learning, will have the effect of changing the fitness landscape. This is part of what is known as the Baldwin effect [7] 1] Gruau and Whitley [5] have found that when evolving Boolean neural networks using grammar trees, genetic search that used learning to change the fitness function was just as effective as Lamarckian strategies: both Lamarckian strategies and search exploiting the Baldwin effect were more efficient and effective than ....

....we will refer to this as a Lamarckian search strategy. In this case, the next state replaces the current state. If the improved solution is merely used to change the fitness of the current state, then the search strategy will be referred to as Baldwinian. Figure 1, taken from Gruau and Whitley [5], illustrates how local search can alter the fitness landscape. Taking N steps deepens the basin of attraction, thus In next ascent, the first improvementfound is taken; if the neighbors of a string are checked in random order, next ascent does not yield a unique solution. For steepest ascent, ....

[Article contains additional citation context not shown here]

F. Gruau and D. Whitley, (1993) Adding learning to the cellular development of neural networks: evolution and the Baldwin effect. Evolutionary Computation 1(3):213-233.

No context found.

Gruau, F. and Whitley, D. (1993) Adding learning to the cellular development of neural networks: Evolution and the Baldwin effect. Evolutionary Computation 1(3):213-233.

No context found.

F. Gruau and D. Whitley. Adding learning to the cellular development of neural networks: Evolution and the baldwin effect. Evolutionary Computation,, 3(1):213--233, MIT Pres, 1993. MIT Press.

No context found.

Frdric Gruau and Darrell Whitley. Adding learning to the cellular development of neural networks: Evolution and the baldwin effect, 1993.

No context found.

Gruau, F. and Whitley, D: Adding learning to the cellular development of neural networks: Evolution and the Baldwin effect, Evolutionary Computation 1(3) (1993), 213--233.

No context found.

F. Gruau and D. Whitley. Adding learning to the cellular development of neural networks: Evolution and the baldwin effect. Evolutionary Computation, 1(3):213-- 234, 1993.

No context found.

F. Gruau, and D. Whitley, "Adding Learning to the Cellular Development of Neural Networks: Evolution and the Baldwin Effect," Evolutionary Computation, Vol. 1, No. 3, 1993, pp. 213--233.

No context found.

: Frederic Gruau and Darrel Whitley: "Adding Learning to the Cellular Development of Neural Networks: Evolution and the Baldwin Effect", in: Evolutionary Computation, No. 1, Vol. 3, pp. 213-233, MIT Press.

No context found.

Whitley, D., and Gruau, F. (1993). Adding learning to the cellular development of neural networks: Evolution and the Baldwin effect. Evolutionary Computation, 1, 213-233.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC