Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374490
G. Pechanek, S. Vassiliadis, J.G. Delgao-Frias
One type of neurocomputer recently proposed, the folded-array digital neural emulator using tree accumulation and communication structures, incorporates a new concept in representing an artificial digital neuron. Beginning from the parallel distributed processing (PDP) neuron model, the folded-array digital neural emulator is briefly described. Then by applying the folded-array concepts to the PDP model, the folded axon/dendrite tree neuron is created which, in a general form, represents a new model for the neural paradigm.<>
{"title":"The folded axon/dendrite tree neuron model","authors":"G. Pechanek, S. Vassiliadis, J.G. Delgao-Frias","doi":"10.1109/ICNN.1994.374490","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374490","url":null,"abstract":"One type of neurocomputer recently proposed, the folded-array digital neural emulator using tree accumulation and communication structures, incorporates a new concept in representing an artificial digital neuron. Beginning from the parallel distributed processing (PDP) neuron model, the folded-array digital neural emulator is briefly described. Then by applying the folded-array concepts to the PDP model, the folded axon/dendrite tree neuron is created which, in a general form, represents a new model for the neural paradigm.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"277 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134281316","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374416
A. Guazzelli, B. de Faria Leao
This paper describes a comparison between fuzzy ARTMAP and combinatorial neural model-CNM neural networks to solve diagnostic problems in medicine. These two different neural networks models were implemented in HYCONES, a tightly coupled hybrid connectionist expert system that integrates frames with neural networks. HYCONES first prototype used the CNM model and was validated for congenital heart diseases diagnoses. In order to assess HYCONES performance with other well-known neural network architectures, HYCONES' CNM networks were replaced by fuzzy ARTMAP networks. This second prototype was submitted to the same validation protocol used in the assessment of the first one. The results of the comparison between CNM and fuzzy ARTMAP and a proposal to modify fuzzy ARTMAP are presented and discussed.<>
{"title":"Incorporating semantics to ART","authors":"A. Guazzelli, B. de Faria Leao","doi":"10.1109/ICNN.1994.374416","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374416","url":null,"abstract":"This paper describes a comparison between fuzzy ARTMAP and combinatorial neural model-CNM neural networks to solve diagnostic problems in medicine. These two different neural networks models were implemented in HYCONES, a tightly coupled hybrid connectionist expert system that integrates frames with neural networks. HYCONES first prototype used the CNM model and was validated for congenital heart diseases diagnoses. In order to assess HYCONES performance with other well-known neural network architectures, HYCONES' CNM networks were replaced by fuzzy ARTMAP networks. This second prototype was submitted to the same validation protocol used in the assessment of the first one. The results of the comparison between CNM and fuzzy ARTMAP and a proposal to modify fuzzy ARTMAP are presented and discussed.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131470280","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374662
B. Baginski, M. Eldracher
A system for robot path planning is presented, that is able to find useful and efficient subgoals in an arbitrary environment. The system consists of two pairs of separately trained networks and an underlying layer of learning units. The network's training is completely based on the most elementary sensoric informations. The created solutions in two and three dimensional simulation environments prove the networks capability to build up a meaningful world model that is effectively applied to the tasks.<>
{"title":"Path planning with neural subgoal search","authors":"B. Baginski, M. Eldracher","doi":"10.1109/ICNN.1994.374662","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374662","url":null,"abstract":"A system for robot path planning is presented, that is able to find useful and efficient subgoals in an arbitrary environment. The system consists of two pairs of separately trained networks and an underlying layer of learning units. The network's training is completely based on the most elementary sensoric informations. The created solutions in two and three dimensional simulation environments prove the networks capability to build up a meaningful world model that is effectively applied to the tasks.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131502559","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.375013
L. Acosta, A. Hamilton, L. Moreno, J.L. Sanchez, J. D. Piñeiro, J. A. Méndez
In this paper we present two methods based on neural networks (NN) for resolution of nonlinear systems optimal control with arbitrary performance index. We have used the minimum time index as an example. Both methods solve the optimal problem for a region of the state space by means of a multistage optimization through a NN chain. Each NN has a fully connected feedforward multilayer structure and the training algorithm for the NN chain is the backpropagation. The chain structure is different for each method, as well as the discretization procedure: classical and block pulse function.<>
{"title":"Two approaches to nonlinear systems optimal control by using neural networks","authors":"L. Acosta, A. Hamilton, L. Moreno, J.L. Sanchez, J. D. Piñeiro, J. A. Méndez","doi":"10.1109/ICNN.1994.375013","DOIUrl":"https://doi.org/10.1109/ICNN.1994.375013","url":null,"abstract":"In this paper we present two methods based on neural networks (NN) for resolution of nonlinear systems optimal control with arbitrary performance index. We have used the minimum time index as an example. Both methods solve the optimal problem for a region of the state space by means of a multistage optimization through a NN chain. Each NN has a fully connected feedforward multilayer structure and the training algorithm for the NN chain is the backpropagation. The chain structure is different for each method, as well as the discretization procedure: classical and block pulse function.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"52 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131132102","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374535
Intaek Kim
A new learning algorithm for multilayer network with bipolar weights (WNBW) is presented. The learning process includes determinations of the bipolar weights of the network and the threshold values at the activation functions in each node. The resultant network performs a perfect recall for given sets of binary input and output pairs. In addition, the network can be easily implemented using digital technology for the realization of its weights.<>
{"title":"Multilayer network with bipolar weights","authors":"Intaek Kim","doi":"10.1109/ICNN.1994.374535","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374535","url":null,"abstract":"A new learning algorithm for multilayer network with bipolar weights (WNBW) is presented. The learning process includes determinations of the bipolar weights of the network and the threshold values at the activation functions in each node. The resultant network performs a perfect recall for given sets of binary input and output pairs. In addition, the network can be easily implemented using digital technology for the realization of its weights.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"49 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128873137","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374409
Fu Li, Jian Zhan
The ART-II model that self-organizes stable recognition codes in real-time is capable of recognizing arbitrary sequences. Based on the feedback mechanism in ART-II, this paper analyses its dynamical process and characteristics of convergence, and defines the concepts of attractive basin, self-stability, focus point. A fuzzy adaptive vigilance /spl rho/ algorithm, with /spl rho/ optimally tailored in signal processing under noisy environment, is proposed. The improved ART-II model with the fuzzy adaptive /spl rho/ has the capability of tolerating and correcting error in the memory while preserving the pattern sensitivity for signal recognition. The new algorithm overcomes the weakness of fixed /spl rho/ which may cause the spurious memory. An intelligent signal processing system is constructed for the recognition of multifrequency patterns in telecommunication. The result of simulation demonstrates that the ART-II model with fuzzy adaptive /spl rho/ recognizes signals at lower signal-to-noise ratio than original one with fixed /spl rho/.<>
{"title":"Fuzzy adapting vigilance parameter of ART-II neural nets","authors":"Fu Li, Jian Zhan","doi":"10.1109/ICNN.1994.374409","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374409","url":null,"abstract":"The ART-II model that self-organizes stable recognition codes in real-time is capable of recognizing arbitrary sequences. Based on the feedback mechanism in ART-II, this paper analyses its dynamical process and characteristics of convergence, and defines the concepts of attractive basin, self-stability, focus point. A fuzzy adaptive vigilance /spl rho/ algorithm, with /spl rho/ optimally tailored in signal processing under noisy environment, is proposed. The improved ART-II model with the fuzzy adaptive /spl rho/ has the capability of tolerating and correcting error in the memory while preserving the pattern sensitivity for signal recognition. The new algorithm overcomes the weakness of fixed /spl rho/ which may cause the spurious memory. An intelligent signal processing system is constructed for the recognition of multifrequency patterns in telecommunication. The result of simulation demonstrates that the ART-II model with fuzzy adaptive /spl rho/ recognizes signals at lower signal-to-noise ratio than original one with fixed /spl rho/.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127824943","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374714
C.-L.J. Hu
When a digitized image is preprocessed by spatial quantizations in a polar-coordinate, the analog vectors representing the r and the /spl theta/ quantizations can be treated separately in neural network trainings. If we apply a segmented Fourier transform (similar to FFT) to the /spl theta/ vector and a segmented Hankel transform to the r vector in a noniterative perceptron training system, then not only the learning of the training patterns is very fast, but also the recognition of an untrained pattern is very robust. Specially the recognition is very robust when the test pattern is rotated even though all the training patterns are not rotated in space. The high robustness of recognition is due to the special preprocessing scheme and the optimum noniterative training scheme we adopted in the design. This paper concentrates at the theoretical origin and the experimental results of the robustness of this novel perceptron learning system.<>
{"title":"Fourier-transformed preprocessing used in a noniteratively-trained perceptron pattern recognizer","authors":"C.-L.J. Hu","doi":"10.1109/ICNN.1994.374714","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374714","url":null,"abstract":"When a digitized image is preprocessed by spatial quantizations in a polar-coordinate, the analog vectors representing the r and the /spl theta/ quantizations can be treated separately in neural network trainings. If we apply a segmented Fourier transform (similar to FFT) to the /spl theta/ vector and a segmented Hankel transform to the r vector in a noniterative perceptron training system, then not only the learning of the training patterns is very fast, but also the recognition of an untrained pattern is very robust. Specially the recognition is very robust when the test pattern is rotated even though all the training patterns are not rotated in space. The high robustness of recognition is due to the special preprocessing scheme and the optimum noniterative training scheme we adopted in the design. This paper concentrates at the theoretical origin and the experimental results of the robustness of this novel perceptron learning system.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127412042","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374752
J. B. Galván, M. J. Pérez-Ilzarbe
The authors have adapted two standard networks, multilayer perceptron and Hopfield, in order to use them for the linear systems identification problem. A systematic method for the study of the order and the delay of the transfer function is explained. Some results using simulated and real data are presented.<>
{"title":"Two neural networks for solving the linear system identification problem","authors":"J. B. Galván, M. J. Pérez-Ilzarbe","doi":"10.1109/ICNN.1994.374752","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374752","url":null,"abstract":"The authors have adapted two standard networks, multilayer perceptron and Hopfield, in order to use them for the linear systems identification problem. A systematic method for the study of the order and the delay of the transfer function is explained. Some results using simulated and real data are presented.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"17 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133752491","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374963
Ren-Jean Liou, M. Azimi-Sadjadi, D. Reinke, T. Vonderhaar, K. E. Eis
This paper presents a neural network-based approach for the detection/classification of cloud field from satellite data in both the visible and infrared (IR) range. Unlike many existing cloud detection schemes which use thresholding and statistical methods, this approach uses singular value decomposition (SVD) to extract image textural features in addition to mean value methodologies. The extracted features are then presented to a self-organizing feature map or Kohonen network for automatic detection and classification of cloud areas. The effectiveness of this method is demonstrated under many situations which are considered difficult for the conventional methods. The proposed method also possesses some interesting classification capabilities which can facilitate future studies on global climatology.<>
{"title":"Detection and classification of cloud data from geostationary satellite using artificial neural networks","authors":"Ren-Jean Liou, M. Azimi-Sadjadi, D. Reinke, T. Vonderhaar, K. E. Eis","doi":"10.1109/ICNN.1994.374963","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374963","url":null,"abstract":"This paper presents a neural network-based approach for the detection/classification of cloud field from satellite data in both the visible and infrared (IR) range. Unlike many existing cloud detection schemes which use thresholding and statistical methods, this approach uses singular value decomposition (SVD) to extract image textural features in addition to mean value methodologies. The extracted features are then presented to a self-organizing feature map or Kohonen network for automatic detection and classification of cloud areas. The effectiveness of this method is demonstrated under many situations which are considered difficult for the conventional methods. The proposed method also possesses some interesting classification capabilities which can facilitate future studies on global climatology.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"171 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115579479","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-06-27DOI: 10.1109/ICNN.1994.374786
C. Posthoff, S. Schawelski, M. Schlosser
The paper shows experiments how to transform knowledge from an endgame database (i.e. a complete collection of information items) into a neural network. In the authors' opinion, it is the first usage of a neural network in the game of chess. Because of complexity it was not possible to deal with the game of chess as a whole, but only with a small endgame. Results and open questions are discussed.<>
{"title":"Neural network learning in a chess endgame","authors":"C. Posthoff, S. Schawelski, M. Schlosser","doi":"10.1109/ICNN.1994.374786","DOIUrl":"https://doi.org/10.1109/ICNN.1994.374786","url":null,"abstract":"The paper shows experiments how to transform knowledge from an endgame database (i.e. a complete collection of information items) into a neural network. In the authors' opinion, it is the first usage of a neural network in the game of chess. Because of complexity it was not possible to deal with the game of chess as a whole, but only with a small endgame. Results and open questions are discussed.<<ETX>>","PeriodicalId":209128,"journal":{"name":"Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94)","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115767224","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：