Xin Yao and Young Liu, "A new evolutionary system for evolving artificial neural networks," IEEE Transactions on Neural Networks, vol. 8, pp. 694--7130, 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....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 ....

X. Yao and Y. Liu. A new evolutionary system for evolving artificial neural networks. IEEE Trans. on Neural Networks, 8(3):694--713, 1997.

....which hidden units when used by each of the ENN set members (with between and 10 hidden units used) 4 Discussion As described in Section 1, a single composite error term cannot be meaningfully propagated through a NN in a multi objective application. The hybrid ENN methods used for example in [56] and highlighted in [55] where indi vidual networks are trained at each generation using a gradient descent technique in addition to their evolutionary manipulation, are also infeasible for these reasons. Instead a general MOENN framework has been introduced in this study where ENN parameters ....

X. Yao and Y. Liu. A New Evolutionary System for Evolving Artificial Neiral Networks. IEEE Transactions on Neural Networks, 8(3):694-713, 1997.

....to use evolutionary search. Genetic algorithms [23,54,56] evolutionary programming [18,20] and evolution strategies [67,69,73] are typical examples of evolutionary search. Attempts at training feedforward neural networks by evolutionary search include the work of Fogel et al. 19] Yao and Liu [87], and Montana and Davis [57] There are also attempts to evolve recurrent networks, e.g. Angeline et al. 3] and McDonnell and Waagen [51] Applying evolutionary search to more complex types of neural networks (high order networks, for example) can be found in [33 35,85] and a good review of ....

....and network weights are changed in a direction opposite to the gradient. This can be computationally expensive if a large number of iterations are required to find an acceptable network. While promising results can be obtained by combining backpropagation and evolutionary search (e.g. in [62,87]) fast variants of backpropagation are sometimes required to speed up the hybrid algorithms. Considering the computational trade offs between local and evolutionary search, Braun and Zagorski [8] adopted a fast backpropagation algorithm RPROP [68] as the local search method. In their hybrid ....

[Article contains additional citation context not shown here]

X. Yao and Y. Liu. A new evolutionary system for evolving artificial neural networks. IEEE Transactions on Neural Networks, 8(3):694--713, 1997.

....the medical task and according to the evolutionary technique. I have chosen to concentrate mainly on articles in archival journals so as to limit the explosive number of references. Table A.1 summarizes these two classifications. A.3.1 According to the medical task 1. Data Mining. a) Diagnosis [7 9, 11, 12, 14, 46, 52, 54, 70, 85, 88, 114, 122, 125, 138, 146, 153, 188]. Diagnosis is the process of selectively gathering information concerning a patient, and interpreting it according to previous knowledge, as evidence for or against the presence or absence of disorders (Section A.2.1) The papers in this category apply evolutionary algorithms to solve numerous ....

.... 174, 176,179,197] 187,189 191] Genetic Multidimensional [28, 30,55] 23,136,192] algorithms Real valued [54] 3,100,156] 38, 57,87] Rule encoding [11,12] 12] Indexed [84] Genetic programming [11, 52,114] Evolution strategies [7 9] 135] Evolutionary programming [46,114,188] Hybrid Evolutionary fuzzy [7 9, 70,125] 86] 176] systems Evolutionary neural [7 9, 46, 122, 188] 22, 25, 33, 72, 73, 98, 110, 166] 24, 31,58] in peripheral blood flow, lymphography classification, brain tumor classification, risk evaluation of heart and coronary ....

[Article contains additional citation context not shown here]

X. Yao and Y. Liu. A new evolutionary system for evolving artificial neural networks. IEEE Transactions on Neural Networks, 8(3):694--713, May 1997.

....Utans, 1994] is used by Cholewo and Zurada (1997) for the construction of instances of time delay and recurrent ANNs for time series prediction [Cholewo and Zurada, 1997] Here, neurons are added sequentially up to a preset limit in order to improve training accuracy and to reduce training time. Yao and Liu (1997) report on time series prediction results using their EPNet system evolving ANN structure and weights simultaneously [Yao and Liu, 1997] Another evolutionary prediction system has been presented by De Falco et al. 1998) where a fixed multi layer perceptron architecture has been trained by ....

....prediction [Cholewo and Zurada, 1997] Here, neurons are added sequentially up to a preset limit in order to improve training accuracy and to reduce training time. Yao and Liu (1997) report on time series prediction results using their EPNet system evolving ANN structure and weights simultaneously [Yao and Liu, 1997]. Another evolutionary prediction system has been presented by De Falco et al. 1998) where a fixed multi layer perceptron architecture has been trained by evolving the weights of the connections [DeFalco et al. 1998] 1.1.1 Training data set selection A less investigated problem concerns the ....

[Article contains additional citation context not shown here]

Yao, X. and Liu, Y. (1997). A New Evolutionary System for Evolving Artificial Neural Networks. IEEE Transactions on Neural Networks, 8(3):694--713.

....their fitness in P (i) Apply reproduction operators to the parents and produce offspring, the next generation, P(i 1) is obtained from the offspring and possibly parents. 18 work of manually finding an optimal network [2] 6] 7] 14] 18] 19] 20] 21] 37] 50] 60] 69] 70] 77] 81] [83] [84] 85] The advantage of the automatic design over the manual design becomes clearer as the complexity of ANN increases. EANNs provide a general framework for investigating various aspects of simulated evolution and learning [10] 14] 15] 50] 52] 3.2.1 General Framework for EANNs In ....

Yao X and Liu Y (1997), A new evolutionary system for evolving artificial neural networks, IEEE Transactions on Neural Networks, 8(3), pp. 694-713.

.... binary [7 9,47,54,57,75,79,83] 93,94,96,105] 60,61,70,89,90] 20,22,39] Multidimensional [17,77,104] 38] Real valued [2,62,85] 25,40,56] Rule encoding [10,11] 11] 53] Indexed [10,36,71] Genetic programming Evolution strategies [7 9] 76] Evolutionary programming [30,71,100] [12] Classifier systems [12,74,80] Hybrid systems [94] 55] 7 9,47,75] Evolutionary fuzzy [16,19,24,48,49,60,70,90] 18,23,41] Evolutionary neural [7 9,30,74,100] C.A. Pena Reyes, M. Sipper Artificial Intelligence in Medicine 19 (2000) 1 23 17 4.1. According to the medical ....

.... [11] 53] Indexed [10,36,71] Genetic programming Evolution strategies [7 9] 76] Evolutionary programming [30,71,100] 12] Classifier systems [12,74,80] Hybrid systems [94] 55] 7 9,47,75] Evolutionary fuzzy [16,19,24,48,49,60,70,90] 18,23,41] Evolutionary neural [7 9,30,74,100] C.A. Pena Reyes, M. Sipper Artificial Intelligence in Medicine 19 (2000) 1 23 17 4.1. According to the medical task 4.1.1. Data mining 4.1.1.1. Diagnosis. Diagnosis is the process of selectively gathering information concerning a patient, and interpreting it according to previous ....

[Article contains additional citation context not shown here]

Yao X, Liu Y. A new evolutionary system for evolving artificial neural networks. IEEE Trans Neural Netw 1997;8(3):694 -- 713.

....Table 4: Comparisons with other works : Audstralian credit approval. Avg Pearson Entropy EPNet Error 0.089 0.092 0.096 0.115 Cal5 ITrule DIPOL92 Discrim Error 0.131 0.137 0.141 0.141 Logdisc CART RBF CASTLE ERror 0.141 0.145 0.145 0.148 NaiveBay IndCART BP Error 0.151 0.152 0. 154 EPNet [16]. FNNCA is an ANN with constructive algorithm and HDANNS is an optimized ANN constructed by hand with trial and error. EPNet is an evolutionarily constructed ANN by Yao and Liu. The comparisons show that the BKS combination of speciated ANNs outperforms FNNCA and EPNet. Although HDANNS shows ....

X. Yao and Y. Liu, "A new evolutionary system for evolving artificial neural networks," IEEE Trans. Neural Networks, vol 8, pp.694-713, Anchorage, USA, 4-9 May 1998.

....others. Other problems decompose into more subtle relationships. Two examples will be examined in this thesis; the multiplexer problem and the robot control problem. This approach to problem solving is not new, and a variety of researchers including Arbib [3, 5, 4] Lewis [66] Brooks [15] Yao [111, 67, 112, 113], and Gruau [32, 33, 34, 37] have used the idea of partial solutions. Each of the above researchers uses a different approach to embed reusable components within a solution. Below is a summary: 1. Arbib: Existing biological systems provide working solutions to difficult problems. Arbib [3, 5, 4] ....

....was augmented the behaviour of the robot improved significantly. The rationale used to describe the need for this type of architecture is that these networks may provide an indication of what is required to automatically build massive networks to carry out complex sensory motor tasks. 4. Yao: Yao [67, 111, 112, 113] uses a modified evolutionary algorithm to create neural networks that generalise well. Yao uses a minimalist evolutionary algorithm that recombines neural modules. The algorithm has the following attributes: a) No crossover: The Yao algorithm does not utilise crossover, mutation is the only ....

X. Yao and Y. Liu. A new evolutionary system for evolving artificial neural networks. Transactions on Neural Networks, 8(3):694--713, May 1997.

No context found.

X. Yao and Y. Liu, "A new evolutionary system for evolving artificial neural networks," IEEE Trans. Neural Networks, vol. 8, pp. 694--713, May 1997.

No context found.

X. Yao and Y. Liu. A new evolutionary system for evolving artificial neural networks. IEEE Trans. on Neural Networks, 8(3):694--713, 1997.

No context found.

X. Yao and Y. Liu, "A new evolutionary system for evolving artificial neural networks," IEEE Transactions on Neural Networks, vol. 8, no. 3, pp. 694--713, 1997.

....with large bias. If we put too much emphasis on learning accuracy, we might get a very big neural network with large variance. EPNet is an alternative method for evolving neural network with good generalization that avoids such a trade off. No network complexity term was introduced in EPNet [4]. In EPNet, a feedforward neural network is evolved using an evolutionary programming algorithm [5] Both the weights and architectures (i.e. connectivity of the network) are evolved in the same evolutionary process. The network may grow as well as shrink. EPNet encourages parsimony of evolved ....

....as follows: Section 2 describes EPNet for evolving neural networks. Section 3 presents the application of EPNet on the Hang Seng stock index data set. Finally, Section 4 concludes with a brief summary of the paper. 2 Evolving Neural Network EPNet evolves generalized feed forward neural networks [4]. It uses direct representation in order to evolve weights in the same evolutionary process as the evolution of network architectures. There is no restriction on the connectivity pattern. Any feed forward architectures are allowed. The main steps of EPNet can be described as follows [4] 1. ....

[Article contains additional citation context not shown here]

X. Yao and Y. Liu. A new evolutionary system for evolving artificial neural networks. IEEE Transactions on Neural Networks, 8(3):694--713, 1997.

....in the ANN field and statistics [2] 3] few attempts have been made in evolutionary learning to use population information in forming the final system. We have carried out experimental studies on three real world problems to demonstrate the effectiveness of our proposed approach. We used EPNet [7] [9] an automatic ANN design tool based on an evolutionary programming algorithm, to evolve a population of ANN s. The performance of the best individual evolved was compared to that of the integrated system which was linearly combined from all the individuals in the last generation. The ....

....system is expected to produce better results. This paper confirms that this is true by conducting a set of computational studies. III. AN EVOLUTIONARY DESIGN SYSTEM FOR ANNS EPNet EPNet is an automatic system based on evolutionary programming (EP) 19] 20] for designing feedforward ANN s [7] [9] The main structure of the system is shown in Fig. 1. EPNet does not use recombination operators in the simulated evolution in order to avoid the permutation (i.e. competing conventions) problem [21] 23] It relies on novel mutations and a rank based selection scheme [24] EPNet evolves ....

[Article contains additional citation context not shown here]

X. Yao and Y. Liu, "A new evolutionary system for evolving artificial neural networks," IEEE Trans. Neural Networks, vol. 8, no. 3, pp. 694--713, 1997.

....TIME SERIES PREDICTION PROBLEM.THE TESTING RMS IN THE TABLE REFERS TO THE ERROR DEFINED BY (30) ON THE TESTING SET. MEAN, SD, MIN, AND MAX INDICATE THE MEAN VALUE, STANDARD DEVIATION,MINIMUM AND MAXIMUM VALUE,RESPECTIVELY TABLE VIII GENERALIZATION RESULTS COMPARISON AMONG CELS, EPNet [16], BP, AND CC LEARNING [17] FOR THE MACKEY GLASS TIME SERIES PREDICTION PROBLEM.THE TESTING RMS IN THE TABLE REFERS TO THE ERROR DEFINED BY (30) ON THE TESTING SET The ensemble architecture used in the experiments was composed of 20 individual networks. Each individual network was a ....

....The correlation penalty term given in (10) was used in this experiment. 2) Experimental Results and Comparisons: Table VII shows the average results of CELS over 25 runs. Each run of CELS was from different initial weights. Table VIII compares CELS s results with those produced by EPNet [16], BP and the cascade correlation (CC) learning [17] It is obvious that CELS was able to achieve the generalization performance better than that of others. For a large time span CELS s results also compared favorably with those produced by Martinetz et al. 15] which had been shown to be better ....

[Article contains additional citation context not shown here]

X. Yao and Y. Liu, "A new evolutionary system for evolving artificial neural networks," IEEE Trans. Neural Networks, vol. 8, pp. 694--713, 1997.

....some important issues that remain open. First, the evolutionary design approach can explore a much wider range of design alternatives than those that could be considered by human beings. This has been shown by many experiments in other design tasks, such as evolutionary design of neural networks [39] [43] building architectures [44] and arts [45] These experiments demonstrated how evolutionary techniques could be applied to evolving novel designs that were difficult to discover by human beings. However, all of these experiments were carried out by software simulation although some of the ....

.... in which evolution is another form of adaptation in addition to learning [71] 25] 27] In particular, EANN s that adapt their architectures through simulated evolution and their weights through learning (training) have been shown to be successful in dealing with a number of benchmark problems [39] [43] 72] Adaptive EHW is closely related to EANN s. For example, both the function level EHW (FEHW) 30] 35] and EPNet [39] 43] evolve feedforward architectures. Both can have different node functions in an architecture [35] 73] However, node functions in FEHW usually have more ....

[Article contains additional citation context not shown here]

X. Yao and Y. Liu, "A new evolutionary system for evolving artificial neural networks," IEEE Trans. Neural Networks, vol. 8, pp. 694--713, May 1997.

....to lower dimensional problems, so the number of hidden nodes can match the number of inputs and explore the re representation issue rather than the dimensionality reduction issue. The choice of networks is an obvious step where an evolutionary algorithm could be implemented. Previous work [11] has evolved neural networks for various tasks, showing exceptional resilience to manipulation of hidden nodes. Evolving the choice and combinations of nodes seems an ideal task for a co evolutionary environment. In this setup an individual would be a certain combination of nodes, and the fitness ....

Xin Yao and Yong Liu. A new evolutionary system for evolving artificial neural networks. IEEE Transactions on Neural Networks, 8(3):694--713, May 1997.

No context found.

Xin Yao and Young Liu, "A new evolutionary system for evolving artificial neural networks," IEEE Transactions on Neural Networks, vol. 8, pp. 694--7130, 1997.

No context found.

Yao, X., Liu, Y. (1997): A New Evolutionary System for Evolving Artificial Neural Networks. IEEE Transaction on Neural Networks 8, 694--713

No context found.

X. Yao and Y. Liu. A New Evolutionary System for Evolving Artificial Neural Networks. IEEE Transactions on Neural Networks, 8(3):694--713, 1997.

No context found.

Yao, X., Liu, Y. (1997): A New Evolutionary System for Evolving Artificial Neural Networks. IEEE Transaction on Neural Networks 8, 694--713

No context found.

X. Yao and Y. Liu, "A New Evolutionary System for Evolving Artificial Neural Networks," IEEE Transactions on Neural Networks, vol. 8, no. 3, pp. 694--713, 1997.

No context found.

X. Yao, Y. Liu, A new evolutionary system for evolving artificial neural networks, IEEE Transactions on Neural Networks 8 (1997) 694--713.

No context found.

X. Yao and Y. Liu. A new evolutionary system for evolving artificial neural networks. IEEE Transactions on Neural Networks, 8(3):694--713, 1997.

No context found.

Xin Yao and Yong Liu. A new evolutionary system for evolving artificial neural networks. IEEE Transactions on Neural Networks, 8(3), May 1997.

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