V. Maniezzo. Genetic evolution of the topology and weight distribution of neural networks. IEEE Transactions on Neural Networks, 5(1), january 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....method derived from the famous optimisation strategies called Genetic algorithms has been proposed. Compared to the Backpropagation training algorithm the Genetic algorithm is quite simple and has a high ability of being parallelized to accelerate the process further. A few papers [HA94] and [Man94] have shown that a training method based on a parallel Genetic algorithm might be a good alternative to the Backpropagation algorithm. The resulting neural networks are as good as the ones trained using the Backpropagation algorithm. Using a parallel machine and a parallel version of the Genetic ....

Vittorio Maniezzo. Genetic evolution of the topology and weight distribution of neural networks. IEEE Transactions on Neural Networks, 5(1):39--53, January 1994.

....and natural representation. The key problem (other than being trapped in a local minimum) with BP and other traditional training algorithms is the choice of a correct architecture (number of hidden nodes and connections) This problem has been tackled by the evolutionary approach in many studies [4, 14, 17, 21, 24, 39, 40, 41]. In some of these studies, weights and architectures are evolved simultaneously. The major disadvantage to the EANN approach is it is computationally expensive, as the evolutionary approach is normally slow. To overcome the slow convergence of the evolutionary approach to ANN, hybrid techniques ....

V. Maniezzo. Genetic evolution of the topology and weight distribution of neural networks. IEEE Transactions on Neural Networks, 5(1):39--53, 1994.

....noisy fitness evaluation problem is to have a one to one mapping between genotypes and phenotypes. That is, both architecture and weight information are encoded in individuals and are evolved simultaneously. Although the idea of evolving both architectures and weights is not new [3] 10] 13] [26], few have explained why it is important in terms of accurate fitness evaluation. The simultaneous evolution of both architectures and weights can be summarized by Fig. 2. The evolution of ANN architectures in general suffers from the permutation problem [27] 28] or called competing conventions ....

V. Maniezzo, "Genetic evolution of the topology and weight distribution of neural networks," IEEE Trans. Neural Networks, vol. 5, pp. 39--53, 1994.

....and natural representation. The key problem (other than being trapped in a local minimum) with BP and other traditional training algorithms is the choice of a correct architecture (number of hidden nodes and connections) This problem has been tackled by the evolutionary approach in many studies [12, 15, 20, 30, 31]. In some of these studies, weights and architectures were evolved simultaneously. The major disadvantage to the EANN approach is it is computationally expensive, as the evolutionary approach is normally slow. To overcome the slow convergence of the evolutionary approach to ANN, hybrid techniques ....

V. Maniezzo. Genetic evolution of the topology and weight distribution of neural networks. IEEE Transactions on Neural Networks, 5(1):39--53, 1994.

.... to use of those evolutionary computation (EC) methods which have previously been applied to uni objective NN design, genetic algorithms (GAs) evolution strategies (ES) and particle swarm optimisation (PSO) GAs have previously be used for feature selection [8, 53] and topography selection [2, 5, 29, 35, 36, 38, 52] and ESs have been used for weight optimisation [21, 42, 45, 55] and adaptive topography selection [15, 37, 57] The recent EC technique of PSO [27] has also proved popular as a uni objective NN optimiser [10, 12, 13, 26, 48] 2 Multi objective evolutionary neural network flamework The use of ....

V. Maniezzo. Genetic Evolution of the Topology and Weight Distribution of Neural Networks. IEEE Trans- actions on Neural Networks, 5(1):39-53, 1994.

....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 ....

....and the differences are summarized in this section. 2.1 Nature of Local Search Local search aims at searching for better solutions in the neighborhood of the current solution. There are different local search methods, and the following are some typical examples. Stochastic Methods Maniezzo [49] proposed a hybrid algorithm for evolving feedforward networks. In Maniezzo s work, the networks are enhanced by a local search method similar to the simplex procedure in linear programming [12] More specifically, the local search is embedded in evolutionary search as a kind of evolutionary ....

V. Maniezzo. Genetic evolution of the topology and weight distribution of neural networks. IEEE Transactions on Neural Networks, 5(1):39--53, 1994. Ku, Mak, and Siu

....approaches have been taken in the direct encoding scheme. The first separates the evolution of architectures from that of connection weights [24] 150] 153] 154] 165] 167] 169] 170] The second approach evolves architectures and connection weights simultaneously [149] 179] 180] [182], 185] 200] This section will focus on the first approach. The second approach will be discussed in Section III D. In the first approach, each connection of an architecture is directly specified by its binary representation [24] 150] 153] 154] 165] 167] 169] 170] 202] For ....

....any weight information has difficulties in evaluating fitness accurately. As a result, the evolution would be very inefficient. One way to alleviate this problem is to evolve ANN architectures and connection weights simultaneously [37] 42] 45] 149] 165] 166] 169] 172] 179] 180] [182], 185] 200] 230] 232] In this case, each individual in a population is a fully specified ANN with complete weight information. Since there is a one to one mapping between a genotype and its phenotype, fitness evaluation is accurate. One issue in evolving ANN s is the choice of search ....

V. Maniezzo, "Genetic evolution of the topology and weight distribution of neural networks," IEEE Trans. Neural Networks, vol. 5, pp. 39--53, Jan. 1994.

....buy and hold strategies. In [40] the modeling of the German stock index DAX is attempted with a neuro fuzzy approach. 2. 2 Neural networks and Evolutionary algorithms Fundamental implementations of neural networks generated and tuned by genetic algorithms can be found in a series of publications [41], 42] 43] 44] 45] 46] The idea behind the implementation of such a hybrid system is the adoption of an evolutionary algorithm for the determination of neural network s weights or the neural network s architecture, or both. In the first case, neural networks are tuned by evolutionary ....

Maniezzo V., Genetic evolution of the topology and weight distribution of neural networks, IEEE Trans. Neural Networks NN 5 39-53, 1994

....the MLP, thus derived using VGA, and its comparison with conventional MLP (having several architectures) are provided for the above mentioned three data sets. In this context we mention other research described in the literature for determining the architecture of neural networks using GAs [5] [6]. In both these approaches, the weights and or the connectivity information are encoded in the chromosomes, and their appropriate values are evolved by the GA. On the other hand, the proposed algorithm determines the appropriate hyperplanes using GAs, and then designs the MLP directly so as to ....

V. Maniezzo, "Genetic evolution of the topology and weight distribution of neural networks," IEEE Trans. Neural Networks, vol. 5, no. 1, pp. 39--53, 1994.

....where [000: 0] decodes to Gamma128 and [111: 1] decodes to 128. An additional bit is assigned to each weight to indicate the presence or absence of the link. If this bit is 0 the remaining bits are unrepresented in the phenotype. The total number of bits in the string is therefore dynamic [9]. Thus a total of 17 bits are assigned for each weight. The fuzzification parameters tuned are the centers (c) and radius ( for each of the linguistic attributes low, medium and high of each feature (eqn. 3) and the output fuzzifiers f d and f e (eqn. 5) These are also coded as 16 bit ....

V. Maniezzo. Genetic evolution of the topology and weight distribution of neural networks. IEEE Transactions on Neural Networks, 5:39--53, 1994.

....scheme is well within the computing power of a standard PC. The micro GA employed is specifically designed to minimize the required number of function evaluations at the upper level. In contrast, using a GA directly to determine the network structure as well as to learn all the network parameters [5] [7] will require far more extensive computation. IV. NONLINAR TIME SERIES APPLICATION Nonlinear time series modeling and prediction are used to illustrate the combined GA and the ROLS learning approach. A. Example 1 This was the simple example of modeling the scalar function (15) used to ....

V. Maniezzo, "Genetic evolution of the topology and weight distribution of neural networks," IEEE Trans. Neural Networks, vol. 5, pp. 39--53, 1994.

.... systems have been developed over the last decade that evolve both neural network topologies and weights (Angeline et al. 1993; Braun and Weisbrod 1993; Dasgupta and McGregor 1992; Fullmer and Miikkulainen 1992; Gruau et al. 1996; Krishnan and Ciesielski 1994; Lee and Kim 1996; Mandischer 1993; Maniezzo 1994; Opitz and Shavlik 1997; Pujol and Poli 1998; Yao and Liu 1996; Zhang and Muhlenbein 1993) These methods encompass a range of ideas about how Topology and Weight Evolving Artificial Neural Networks (TWEANNs) should be implemented. In this section, we address some of the ideas and assumptions ....

....one. Direct encoding schemes, employed by most TWEANNs, specify in the genome every connection and node that will appear in the phenotype (Angeline et al. 1993; Braun and Weisbrod 1993; Dasgupta and McGregor 1992; Fullmer and Miikkulainen 1992; Krishnan and Ciesielski 1994; Lee and Kim 1996; Maniezzo 1994; Opitz and Shavlik 1997; Pujol and Poli 1998; Yao and Liu 1996; Zhang and Muhlenbein 1993) In contrast, indirect encodings usually only specify rules for constructing a phenotype (Gruau 1993; Mandischer 1993) These rules can be layer specifications or growth rules through cell division. ....

Maniezzo, V. (1994). Genetic evolution of the topology and weight distribution of neural networks. IEEE Transactions on Neural Networks, 5(1):39--53.

....for these parameters [13] Still, it is not at all clear, if these parameters are robust or if they depend on the specific problems. In a few of the GANN systems, some of these parameters are encoded in the genome, for instance: the learning rate in [16] or number of encoding bits per weight in [18]. This encoding solves part of the problem at the cost of an increase of the search space and thus the convergence time. Comparison of encoding techniques The inexistence of a theoretical account makes the comparison of different GANN systems difficult. The empirical results, however, do not ....

Vittorio Maniezzo. Genetic evolution of the topology and weight distribution of neural networks. IEEE Transactions of Neural Networks, 5(1):39-- 53, 1994.

....are several This work was supported by the Thuringian Ministry for Science, Research, and Arts (Project ITHERA) pruning algorithms (e.g. CFP97, Set97] which iteratively reduce the network s complexity. Evolutionary optimization of the network topology has also been investigated extensively [YL97, Man94, BZ94]. In this paper, we investigate a technique using the A Algorithm (A ) for nding optimal network topologies. This new idea was proposed in [DGW97] It was proved that under certain conditions, the A nds the smallest network topology for a given pattern set in minimal time. The A searches a ....

V. Maniezzo. Genetic Evolution of the Topology and Weight Distribution of Neural Networks. IEEE Transactions on Neural Networks, 5(1):39-53, January 1994.

....are several This work was supported by the Thuringian Ministry for Science, Research, and Arts (Project ITHERA) pruning algorithms (e.g. CFP97, Set97] which iteratively reduce the network s complexity. Evolutionary optimization of the network topology has also been investigated extensively [YL97, Man94, BZ94]. In this paper, we investigate a technique using the A Algorithm (A ) for finding optimal network topologies. This new idea was proposed in [DGW97] It was proved that under certain conditions, the A finds the smallest network topology for a given pattern set in minimal time. The A searches a ....

V. Maniezzo. Genetic Evolution of the Topology and Weight Distribution of Neural Networks. IEEE Transactions on Neural Networks, 5(1):39--53, January 1994.

.... three binary strings was already shown to compare favourably with standard crossover on the DeJong Benchmark functions [ Better performance resulting from the use of Simplex GA for binary strings in the context of the classical neural network architecture discovery problem are also presented in [12]. To keep this idea of a repulsive effect exerted by the third string, the genes taken from this worst parent were just inverted before to be integrated in the offspring. The problem with the TSP is that there is no notion of opposite for an edge, and thus a new way of expressing the repulsive ....

Maniezzo, V., 1994. "Genetic Evolution of the Topology and Weight Distribution of Neural Networks" - in IEEE Transactions on Neural Networks - Vol.5, No1 - pp. 39-53.

.... on the neural network directly, and rely exclusively on mutation [10, 11, 12, 13] or combine mutation with training [14] Methods based on genetic algorithms usually represent the structure and the weights of ANNs as a string of 1 bits or as a combination of bits, integers and real numbers [15, 16, 17, 18, 19, 20], and perform the crossover operation as if the network were a linear structure. However, neural networks cannot naturally be represented as vectors. They are oriented graphs, whose nodes are neurons and whose arcs are synaptic connections. Therefore, it is arguable that any efficient approach to ....

V. Maniezzo. Genetic evolution of the topology and weight distribution of neural networks. IEEE Transactions on Neural Networks, 5(1):39--53, Jan. 1994.

....approach to the development of neural networks [6] it misses the exchange of genetic material which is the driving force in GAs. Methods based on genetic algorithms [4] usually represent the structure and the weights of NNs as a string of bits or as a combination of bits and real numbers [7, 8, 9], and perform the crossover operation as if the network were a linear structure. However, neural networks cannot naturally be represented as binary vectors. They are oriented graphs, whose nodes are neurons and whose arcs are synaptic connections. Therefore, it is arguable that any efficient ....

V. Maniezzo. Genetic evolution of the topology and weight distribution of neural networks. IEEE Transactions on Neural Networks, 5(1):39--53, Jan. 1994.

....a more suitable approach to the development of neural networks, it misses the exchange of genetic material which is the driving force in GAs. Methods based on genetic algorithms usually represent the structure and the weights of NNs as a string of bits or as a combination of bits and real numbers [7, 8, 9], and perform the crossover operation as if the network were a linear structure. However, neural networks cannot naturally be represented as binary vectors. They are oriented graphs, whose nodes are neurons and whose arcs are 1 synaptic connections. Therefore, it is arguable that any efficient ....

V. Maniezzo. Genetic evolution of the topology and weight distribution of neural networks. IEEE Transactions on Neural Networks, 5(1):39--53, Jan. 1994.

No context found.

V. Maniezzo. Genetic evolution of the topology and weight distribution of neural networks. IEEE Transactions on Neural Networks, 5(1), january 1994.

No context found.

V. Maniezzo 1994, "Genetic evolution of the topology and weight distribution of neural networks," IEEE Trans. On Neural Networks 5(1), 39--53.

No context found.

V. Maniezzo, "Genetic Evolution of the Topology and Weight Distribution of Neural Networks," IEEE Transactions on Neural Networks, vol. 5, no. 1, pp. 39--53, 1994. 17

No context found.

V. Maniezzo, "Genetic evolution of the topology and weight distribution of neural networks," IEEE Trans. on Neural Networks, vol. 5, no. 1, pp. 39--53, 1994.

No context found.

V. Maniezzo. Genetic Evolution of the Topology and Weight Distribution of Neural Networks. IEEE Transactions on Neural Networks, 5(1):39--53, 1993.

No context found.

V. Maniezzo, Genetic evolution of the topology and weight distribution of neural networks, in IEEE Transactions on Neural Networks, vol. 5, pp. 3953, 1994.

First 50 documents  Next 50

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