Pub Date : 1991-11-18DOI: 10.1109/IJCNN.1991.170431
J. Coughlin, R. Baran
Backpropagation networks with a single hidden layer were trained to perform one-step prediction on a variety of scalar time series. The performance of such nets typically equals or exceeds that of the linear adaptive predictor of the same order. Comparisons of the linear and nonlinear predictors were made with periodic, chaotic, and random time series, including broadband ocean acoustic ambient noise.<>
{"title":"Time series prediction with linear and nonlinear adaptive networks","authors":"J. Coughlin, R. Baran","doi":"10.1109/IJCNN.1991.170431","DOIUrl":"https://doi.org/10.1109/IJCNN.1991.170431","url":null,"abstract":"Backpropagation networks with a single hidden layer were trained to perform one-step prediction on a variety of scalar time series. The performance of such nets typically equals or exceeds that of the linear adaptive predictor of the same order. Comparisons of the linear and nonlinear predictors were made with periodic, chaotic, and random time series, including broadband ocean acoustic ambient noise.<<ETX>>","PeriodicalId":211135,"journal":{"name":"[Proceedings] 1991 IEEE International Joint Conference on Neural Networks","volume":"99 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1991-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131971493","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 : 1991-11-18DOI: 10.1109/IJCNN.1991.170520
Shigeo Abe
For the extended sigmoid function which is monotonic and differentiable at any interior point in the output range, the author clarifies the condition that a vertex of a hypercube becomes a local minimum of the Hopfield neural networks and a monotonic convergence region to that minimum. Based on this, a method of analyzing and suppressing spurious states in the networks is derived. It is shown that all the spurious states of the traveling salesman problem for the Hopfield original energy function can be suppressed by the proposed method, and its validity is demonstrated by computer simulations.<>
{"title":"Global convergence and suppression of spurious states of the Hopfield neural networks","authors":"Shigeo Abe","doi":"10.1109/IJCNN.1991.170520","DOIUrl":"https://doi.org/10.1109/IJCNN.1991.170520","url":null,"abstract":"For the extended sigmoid function which is monotonic and differentiable at any interior point in the output range, the author clarifies the condition that a vertex of a hypercube becomes a local minimum of the Hopfield neural networks and a monotonic convergence region to that minimum. Based on this, a method of analyzing and suppressing spurious states in the networks is derived. It is shown that all the spurious states of the traveling salesman problem for the Hopfield original energy function can be suppressed by the proposed method, and its validity is demonstrated by computer simulations.<<ETX>>","PeriodicalId":211135,"journal":{"name":"[Proceedings] 1991 IEEE International Joint Conference on Neural Networks","volume":"45 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1991-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133909357","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 : 1991-11-18DOI: 10.1109/IJCNN.1991.170781
M.-C. Lu, Chong-Huah Lo, H. Don
A multistage concurrently processing artificial neural network is proposed to identify 3D unoccluded objects from arbitrary viewing angles and to estimate their poses. 3D moment invariants are used to generate feature vectors from 2-1/2D range images. Objects are recognized via moment invariants which are invariant to translation, scaling, and rotation. The proposed network is divided into two stages, the feature extraction stage and the feature detection stage, to generate moment invariants and detect the input features, respectively. Experimental results show that objects coded by 3D moment invariant features can always be satisfactorily classified and estimated by the proposed neural network.<>
{"title":"A neural network approach to 3D object identification and pose estimation","authors":"M.-C. Lu, Chong-Huah Lo, H. Don","doi":"10.1109/IJCNN.1991.170781","DOIUrl":"https://doi.org/10.1109/IJCNN.1991.170781","url":null,"abstract":"A multistage concurrently processing artificial neural network is proposed to identify 3D unoccluded objects from arbitrary viewing angles and to estimate their poses. 3D moment invariants are used to generate feature vectors from 2-1/2D range images. Objects are recognized via moment invariants which are invariant to translation, scaling, and rotation. The proposed network is divided into two stages, the feature extraction stage and the feature detection stage, to generate moment invariants and detect the input features, respectively. Experimental results show that objects coded by 3D moment invariant features can always be satisfactorily classified and estimated by the proposed neural network.<<ETX>>","PeriodicalId":211135,"journal":{"name":"[Proceedings] 1991 IEEE International Joint Conference on Neural Networks","volume":"69 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1991-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133405247","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 : 1991-11-18DOI: 10.1109/IJCNN.1991.170576
C.D. Wang, J. P. Thompson
Based on a self-organized, probabilistic neural network (PNN) paradigm, a parallel network can be used to sort data parameters into classes with high sorting accuracy and low fragmentation. The capabilities of the sorter, as applied to ESM (electronic support measure) pulse-data sorting, are shown. The PNN implements the statistical Bayesian strategy by computing a joint probability density over all input data parameters to match a group of candidate data classes. The sorting is accomplished by assigning then inputs to the most likely group with highest probability density estimate. Based on test data from an ESM system, the PNN has shown significant improvement over conventional rule-based techniques. The parallel computer architecture of PNN is well-suited for VLSI chip implementation. An 80000 gate semicustom chip design concept is described.<>
{"title":"An adaptive data sorter based on probabilistic neural networks","authors":"C.D. Wang, J. P. Thompson","doi":"10.1109/IJCNN.1991.170576","DOIUrl":"https://doi.org/10.1109/IJCNN.1991.170576","url":null,"abstract":"Based on a self-organized, probabilistic neural network (PNN) paradigm, a parallel network can be used to sort data parameters into classes with high sorting accuracy and low fragmentation. The capabilities of the sorter, as applied to ESM (electronic support measure) pulse-data sorting, are shown. The PNN implements the statistical Bayesian strategy by computing a joint probability density over all input data parameters to match a group of candidate data classes. The sorting is accomplished by assigning then inputs to the most likely group with highest probability density estimate. Based on test data from an ESM system, the PNN has shown significant improvement over conventional rule-based techniques. The parallel computer architecture of PNN is well-suited for VLSI chip implementation. An 80000 gate semicustom chip design concept is described.<<ETX>>","PeriodicalId":211135,"journal":{"name":"[Proceedings] 1991 IEEE International Joint Conference on Neural Networks","volume":"40 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1991-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132590433","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 : 1991-11-18DOI: 10.1109/IJCNN.1991.170527
W. J. Ho, C. Osborne
The authors trained the multi-layered backpropagation neural network to segment two paper samples with very similar paper formation characteristics. The paper samples were chosen deliberately in order to evaluate the multi-layered backpropagation performance in a difficult classification problem. The authors used the texture features obtained from the spatial gray-tone dependence cooccurrence matrices as inputs to the multi-layered backpropagation network. Results show good classification percentages when compared to a subjective evaluation method.<>
{"title":"Texture segmentation using multi-layered backpropagation","authors":"W. J. Ho, C. Osborne","doi":"10.1109/IJCNN.1991.170527","DOIUrl":"https://doi.org/10.1109/IJCNN.1991.170527","url":null,"abstract":"The authors trained the multi-layered backpropagation neural network to segment two paper samples with very similar paper formation characteristics. The paper samples were chosen deliberately in order to evaluate the multi-layered backpropagation performance in a difficult classification problem. The authors used the texture features obtained from the spatial gray-tone dependence cooccurrence matrices as inputs to the multi-layered backpropagation network. Results show good classification percentages when compared to a subjective evaluation method.<<ETX>>","PeriodicalId":211135,"journal":{"name":"[Proceedings] 1991 IEEE International Joint Conference on Neural Networks","volume":"21 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1991-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132891211","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 : 1991-11-18DOI: 10.1109/IJCNN.1991.170503
C.-C. Chen
The author identifies several different neural network models which are related to nearest neighbor learning. They include radial basis functions, sparse distributed memory, and localized receptive fields. One way to improve the neural networks' performance is by using the cooperation of different learning algorithms. The prediction of chaotic time series is used as an example to show how nearest neighbor learning can be employed to improve Sanger's tree-structured algorithm which predicts future values of the Mackey-Glass differential delay equation.<>
{"title":"Using nearest neighbor learning to improve Sanger's tree-structured algorithm","authors":"C.-C. Chen","doi":"10.1109/IJCNN.1991.170503","DOIUrl":"https://doi.org/10.1109/IJCNN.1991.170503","url":null,"abstract":"The author identifies several different neural network models which are related to nearest neighbor learning. They include radial basis functions, sparse distributed memory, and localized receptive fields. One way to improve the neural networks' performance is by using the cooperation of different learning algorithms. The prediction of chaotic time series is used as an example to show how nearest neighbor learning can be employed to improve Sanger's tree-structured algorithm which predicts future values of the Mackey-Glass differential delay equation.<<ETX>>","PeriodicalId":211135,"journal":{"name":"[Proceedings] 1991 IEEE International Joint Conference on Neural Networks","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1991-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133705089","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 : 1991-11-18DOI: 10.1109/IJCNN.1991.170377
S. Chien, In-Cheol Kim, Dae-Young Kim
Autoassociative single-layer neural networks (SLNNs) and multilayer perceptron (MLP) models have been designed to achieve English-character image recall and classification. These two models are trained on the pseudoinverse algorithm and backpropagation learning algorithms, respectively. Improvements on the error-correcting effect of these two models can be achieved by introducing a feedback structure which returns autoassociative image outputs and classification tag fields into the network's inputs. The two models are compared in terms of character image recall and classification capabilities. Experimental results indicative that the MLP network required longer learning time and a smaller number of weights, and showed more stable variations in noise-correcting capability and classification rate with respect to the change of the numbers of stored patterns than the SLNN.<>
{"title":"Iterative autoassociative memory models for image recalls and pattern classifications","authors":"S. Chien, In-Cheol Kim, Dae-Young Kim","doi":"10.1109/IJCNN.1991.170377","DOIUrl":"https://doi.org/10.1109/IJCNN.1991.170377","url":null,"abstract":"Autoassociative single-layer neural networks (SLNNs) and multilayer perceptron (MLP) models have been designed to achieve English-character image recall and classification. These two models are trained on the pseudoinverse algorithm and backpropagation learning algorithms, respectively. Improvements on the error-correcting effect of these two models can be achieved by introducing a feedback structure which returns autoassociative image outputs and classification tag fields into the network's inputs. The two models are compared in terms of character image recall and classification capabilities. Experimental results indicative that the MLP network required longer learning time and a smaller number of weights, and showed more stable variations in noise-correcting capability and classification rate with respect to the change of the numbers of stored patterns than the SLNN.<<ETX>>","PeriodicalId":211135,"journal":{"name":"[Proceedings] 1991 IEEE International Joint Conference on Neural Networks","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1991-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134002750","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 : 1991-11-18DOI: 10.1109/IJCNN.1991.170606
G. Martinelli, R. Prefetti
A novel associative memory model is presented, which is derived from the Hopfield discrete neural network. Its architecture is greatly simplified because the number of interconnections grows only linearly with the dimensionality of the stored patterns. It makes use of a modified backpropagation algorithm as a learning tool. During the retrieval phase the network operates as an autoassociative BAM (directional associative memory), which searches for a minimum of an appropriate energy function. Computer simulations point out the good performances of the proposed learning method in terms of capacity and number of spurious stable states.<>
{"title":"Hopfield network with O(N) complexity using a constrained backpropagation learning","authors":"G. Martinelli, R. Prefetti","doi":"10.1109/IJCNN.1991.170606","DOIUrl":"https://doi.org/10.1109/IJCNN.1991.170606","url":null,"abstract":"A novel associative memory model is presented, which is derived from the Hopfield discrete neural network. Its architecture is greatly simplified because the number of interconnections grows only linearly with the dimensionality of the stored patterns. It makes use of a modified backpropagation algorithm as a learning tool. During the retrieval phase the network operates as an autoassociative BAM (directional associative memory), which searches for a minimum of an appropriate energy function. Computer simulations point out the good performances of the proposed learning method in terms of capacity and number of spurious stable states.<<ETX>>","PeriodicalId":211135,"journal":{"name":"[Proceedings] 1991 IEEE International Joint Conference on Neural Networks","volume":"509 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1991-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122759997","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 : 1991-11-18DOI: 10.1109/IJCNN.1991.170552
P. Vanhoutte, K. Deegan, K. Khorasani
The utilization of a two-stage neural network architecture for the detection of targets in a passive, listen-only sonar is discussed. The two-stage network consists of a first-stage Hopfield network to suppress noise, and a second stage using a bidirectional associative memory (BAM) to make the decision as to whether a target has been detected or not. A second architecture using only a single BAM stage is also presented for illustrative purposes. The target is assumed to be emitting a single tone sinusoid as its signature. The system also assumes only white Gaussian noise perturbation to the signal. It is shown that this network structure provides correct detection at a signal-to-noise ratio of -21 dB, a 6 dB improvement in target detection over a similar network using a perceptron in the second stage. Performance is shown to be limited to the size of the Hopfield network, in the first stage, and to the training set applied to it.<>
{"title":"Passive sonar processing using neural networks","authors":"P. Vanhoutte, K. Deegan, K. Khorasani","doi":"10.1109/IJCNN.1991.170552","DOIUrl":"https://doi.org/10.1109/IJCNN.1991.170552","url":null,"abstract":"The utilization of a two-stage neural network architecture for the detection of targets in a passive, listen-only sonar is discussed. The two-stage network consists of a first-stage Hopfield network to suppress noise, and a second stage using a bidirectional associative memory (BAM) to make the decision as to whether a target has been detected or not. A second architecture using only a single BAM stage is also presented for illustrative purposes. The target is assumed to be emitting a single tone sinusoid as its signature. The system also assumes only white Gaussian noise perturbation to the signal. It is shown that this network structure provides correct detection at a signal-to-noise ratio of -21 dB, a 6 dB improvement in target detection over a similar network using a perceptron in the second stage. Performance is shown to be limited to the size of the Hopfield network, in the first stage, and to the training set applied to it.<<ETX>>","PeriodicalId":211135,"journal":{"name":"[Proceedings] 1991 IEEE International Joint Conference on Neural Networks","volume":"65 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1991-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121842021","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 : 1991-11-18DOI: 10.1109/IJCNN.1991.170404
P. Mills, Albert Y. Zomaya
The authors introduce three aspects of the neural identification of nonlinear systems. First, a method of extending the error backpropagation neural network to enable it to perform online identification of a system is considered. This enables the investigation of adaptive nonlinear process control based on neural identification. Second, the neural identification has been successfully tested on a complex nonlinear composite system which includes formidable, but realistic, nonlinear process characteristics such as hysteresis. This has helped to demonstrate the general applicability of identification using neural techniques. Third, the novel method of neural identification was compared with online identification based on the well-established linear least-squares technique. The comparison highlights the faster adaptation of linear identification against the higher asymptotic accuracy of neural identification.<>
{"title":"A neural network approach to on-line identification of non-linear systems","authors":"P. Mills, Albert Y. Zomaya","doi":"10.1109/IJCNN.1991.170404","DOIUrl":"https://doi.org/10.1109/IJCNN.1991.170404","url":null,"abstract":"The authors introduce three aspects of the neural identification of nonlinear systems. First, a method of extending the error backpropagation neural network to enable it to perform online identification of a system is considered. This enables the investigation of adaptive nonlinear process control based on neural identification. Second, the neural identification has been successfully tested on a complex nonlinear composite system which includes formidable, but realistic, nonlinear process characteristics such as hysteresis. This has helped to demonstrate the general applicability of identification using neural techniques. Third, the novel method of neural identification was compared with online identification based on the well-established linear least-squares technique. The comparison highlights the faster adaptation of linear identification against the higher asymptotic accuracy of neural identification.<<ETX>>","PeriodicalId":211135,"journal":{"name":"[Proceedings] 1991 IEEE International Joint Conference on Neural Networks","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1991-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121299436","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：