Pub Date : 1992-06-07DOI: 10.1109/IJCNN.1992.226872
B. Yuhas
The author examines how well some proposed localization models are able to locate speech stimuli on the azimuth. Working with recordings obtained using a manikin in a modestly reverberant conference room, a variety of existing models in software are evaluated and their results are compared. A model is then proposed, which uses the natural time delays of the cochlea combined with adaptation to obtain instantaneous estimates of position. A system for binaural localization is proposed and its performance is compared to existing models of auditory localization and to methods of direct calculation.<>
{"title":"Automated sound localization through adaptation","authors":"B. Yuhas","doi":"10.1109/IJCNN.1992.226872","DOIUrl":"https://doi.org/10.1109/IJCNN.1992.226872","url":null,"abstract":"The author examines how well some proposed localization models are able to locate speech stimuli on the azimuth. Working with recordings obtained using a manikin in a modestly reverberant conference room, a variety of existing models in software are evaluated and their results are compared. A model is then proposed, which uses the natural time delays of the cochlea combined with adaptation to obtain instantaneous estimates of position. A system for binaural localization is proposed and its performance is compared to existing models of auditory localization and to methods of direct calculation.<<ETX>>","PeriodicalId":286849,"journal":{"name":"[Proceedings 1992] IJCNN International Joint Conference on Neural Networks","volume":"83 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128600067","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 : 1992-06-07DOI: 10.1109/IJCNN.1992.227173
C. Yao, A. Willson
It is shown how a parallel carry generator circuit using the sigmoidal (input/output) characteristic of a neuron can be employed in a carry select adder architecture. The circuit performs the carry generation function in parallel with the generation of the summation bits. By examining the input-output pairs of a digital adder it is found that the generation of its output carry is a most basic mapping of a neural network, the mapping of a single neuron. The realization of this mapping by a transistor circuit is described. Performance results derived from SPICE simulations of the proposed circuit, using 1.2- mu m CMOS technology, are also given.<>
它展示了如何利用神经元的s型(输入/输出)特性的并行进位发生器电路可以用于进位选择加法器结构。该电路与求和位的生成并行地执行进位生成功能。通过研究数字加法器的输入输出对，发现其输出进位的生成是神经网络最基本的映射，即单个神经元的映射。文中描述了用晶体管电路实现这种映射的方法。本文还给出了采用1.2 μ m CMOS技术的电路的SPICE仿真结果。
{"title":"One-neuron circuitry for carry generation in a 4-bit adder","authors":"C. Yao, A. Willson","doi":"10.1109/IJCNN.1992.227173","DOIUrl":"https://doi.org/10.1109/IJCNN.1992.227173","url":null,"abstract":"It is shown how a parallel carry generator circuit using the sigmoidal (input/output) characteristic of a neuron can be employed in a carry select adder architecture. The circuit performs the carry generation function in parallel with the generation of the summation bits. By examining the input-output pairs of a digital adder it is found that the generation of its output carry is a most basic mapping of a neural network, the mapping of a single neuron. The realization of this mapping by a transistor circuit is described. Performance results derived from SPICE simulations of the proposed circuit, using 1.2- mu m CMOS technology, are also given.<<ETX>>","PeriodicalId":286849,"journal":{"name":"[Proceedings 1992] IJCNN International Joint Conference on Neural Networks","volume":"59 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128635215","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 : 1992-06-07DOI: 10.1109/IJCNN.1992.227007
M. Shirazi
It is common practice to store patterns in associative memories by encoding them into vectors with components having binary values 0, +1 or -1, +1. An encoding scheme is said to be sparse if the number of 0's or -1's of the encoding vectors is very large compared to the number of +1's. An asymptotically sparsely encoded associative memory is considered. Patterns are encoded by vectors with components having the values of -1 or +1. The encoding vectors are random realizations of a sequence of n Bernoulli trials heavily biased toward -1. The encoded patterns are stored in the network according to the Hebbian rule. It is proved that the associated crosstalks are asymptotically Gaussian.<>
{"title":"On the crosstalks in sparsely encoded associative memories","authors":"M. Shirazi","doi":"10.1109/IJCNN.1992.227007","DOIUrl":"https://doi.org/10.1109/IJCNN.1992.227007","url":null,"abstract":"It is common practice to store patterns in associative memories by encoding them into vectors with components having binary values 0, +1 or -1, +1. An encoding scheme is said to be sparse if the number of 0's or -1's of the encoding vectors is very large compared to the number of +1's. An asymptotically sparsely encoded associative memory is considered. Patterns are encoded by vectors with components having the values of -1 or +1. The encoding vectors are random realizations of a sequence of n Bernoulli trials heavily biased toward -1. The encoded patterns are stored in the network according to the Hebbian rule. It is proved that the associated crosstalks are asymptotically Gaussian.<<ETX>>","PeriodicalId":286849,"journal":{"name":"[Proceedings 1992] IJCNN International Joint Conference on Neural Networks","volume":"84 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127536515","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 : 1992-06-07DOI: 10.1109/IJCNN.1992.226955
E. Littmann, H. Ritter
A novel incremental cascade network architecture based on error minimization is presented. The properties of this and related cascade architectures are discussed, and the influence of the objective function is investigated. The performance of the network is achieved by several layers of nonlinear units that are trained in a strictly feedforward manner and one after the other. Nonlinearity is generated by using sigmoid units and, optionally, additional powers of their activity values. Extensive benchmarking results for the XOR problem are reported, as are various classification tasks, and time series prediction. These are compared to other results reported in the literature. Direct cascading is proposed as promising approach to introducing context information in the approximation process.<>
{"title":"Cascade network architectures","authors":"E. Littmann, H. Ritter","doi":"10.1109/IJCNN.1992.226955","DOIUrl":"https://doi.org/10.1109/IJCNN.1992.226955","url":null,"abstract":"A novel incremental cascade network architecture based on error minimization is presented. The properties of this and related cascade architectures are discussed, and the influence of the objective function is investigated. The performance of the network is achieved by several layers of nonlinear units that are trained in a strictly feedforward manner and one after the other. Nonlinearity is generated by using sigmoid units and, optionally, additional powers of their activity values. Extensive benchmarking results for the XOR problem are reported, as are various classification tasks, and time series prediction. These are compared to other results reported in the literature. Direct cascading is proposed as promising approach to introducing context information in the approximation process.<<ETX>>","PeriodicalId":286849,"journal":{"name":"[Proceedings 1992] IJCNN International Joint Conference on Neural Networks","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128970495","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 : 1992-06-07DOI: 10.1109/IJCNN.1992.226994
David G. Stork, G. Wolff, Earl Levine
A modified time-delay neural network (TDNN) has been designed to perform both automatic lipreading (speech reading) in conjunction with acoustic speech recognition in order to improve recognition both in silent environments as well as in the presence of acoustic noise. The system is far more robust to acoustic noise and verbal distractors than is a system not incorporating visual information. Specifically, in the presence of high-amplitude pink noise, the low recognition rate in the acoustic only system (43%) is raised to 75% by the incorporation of visual information. The system responds to (artificial) conflicting cross-modal patterns in a way closely analogous to the McGurk effect in humans. The power of neural techniques is demonstrated in several difficult domains: pattern recognition; sensory integration; and distributed approaches toward 'rule-based' (linguistic-phonological) processing.<>
{"title":"Neural network lipreading system for improved speech recognition","authors":"David G. Stork, G. Wolff, Earl Levine","doi":"10.1109/IJCNN.1992.226994","DOIUrl":"https://doi.org/10.1109/IJCNN.1992.226994","url":null,"abstract":"A modified time-delay neural network (TDNN) has been designed to perform both automatic lipreading (speech reading) in conjunction with acoustic speech recognition in order to improve recognition both in silent environments as well as in the presence of acoustic noise. The system is far more robust to acoustic noise and verbal distractors than is a system not incorporating visual information. Specifically, in the presence of high-amplitude pink noise, the low recognition rate in the acoustic only system (43%) is raised to 75% by the incorporation of visual information. The system responds to (artificial) conflicting cross-modal patterns in a way closely analogous to the McGurk effect in humans. The power of neural techniques is demonstrated in several difficult domains: pattern recognition; sensory integration; and distributed approaches toward 'rule-based' (linguistic-phonological) processing.<<ETX>>","PeriodicalId":286849,"journal":{"name":"[Proceedings 1992] IJCNN International Joint Conference on Neural Networks","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130688771","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 : 1992-06-07DOI: 10.1109/IJCNN.1992.227053
M. Idesawa
Two types of occlusion cues in binocular fusion were investigated by using the phenomenon of 3-D illusion. In the first type, the visibility of the occluded object was changed at the border of the contours of the occluding object. In the other type, the visibility of the occluded object was changed at the surface of the occluding object. Here, the visibility was changed when the occluded object passed through the surface from the outside space to the inside space of the occluding object. These occlusion cues have close relations with the visual perception of 3-D space in binocular viewing and can reveal the mechanism underlying the 3-D space perception ability of the human visual system. Based on these occlusion cues, a new interpolation becomes possible for the perception of the random dot stereogram.<>
{"title":"Two types of occlusion cue in 3-D perception with binocular viewing","authors":"M. Idesawa","doi":"10.1109/IJCNN.1992.227053","DOIUrl":"https://doi.org/10.1109/IJCNN.1992.227053","url":null,"abstract":"Two types of occlusion cues in binocular fusion were investigated by using the phenomenon of 3-D illusion. In the first type, the visibility of the occluded object was changed at the border of the contours of the occluding object. In the other type, the visibility of the occluded object was changed at the surface of the occluding object. Here, the visibility was changed when the occluded object passed through the surface from the outside space to the inside space of the occluding object. These occlusion cues have close relations with the visual perception of 3-D space in binocular viewing and can reveal the mechanism underlying the 3-D space perception ability of the human visual system. Based on these occlusion cues, a new interpolation becomes possible for the perception of the random dot stereogram.<<ETX>>","PeriodicalId":286849,"journal":{"name":"[Proceedings 1992] IJCNN International Joint Conference on Neural Networks","volume":"50 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132450399","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 : 1992-06-07DOI: 10.1109/IJCNN.1992.226954
M. Hassoun, C. Wang, A. Spitzer
A signal decomposition method which utilizes a multi-layer dynamic network to automatically decompose a clinical electromyogram (EMG), without supervision, is proposed. Due to the lack of a priori knowledge of motor unit potential (MUP) morphology, the EMG decomposition must be performed in an unsupervised manner. A neural network classifier, consisting of a multi-layer neural net of perceptrons and using an unsupervised training strategy, is proposed. The neural network learns repetitive appearances of MUP waveforms from their suspected occurrence in a given filtered EMLG signal by using an unsupervised clustering strategy. Upon training, the network creates stable attractors which correspond to nominal representations of MUP clusters hidden in the data. The decomposition/clustering capabilities of the proposed method are validated on a real EMG signal and on an unlabeled signal set.<>
{"title":"Electromyogram decomposition via unsupervised dynamic multi-layer neural network","authors":"M. Hassoun, C. Wang, A. Spitzer","doi":"10.1109/IJCNN.1992.226954","DOIUrl":"https://doi.org/10.1109/IJCNN.1992.226954","url":null,"abstract":"A signal decomposition method which utilizes a multi-layer dynamic network to automatically decompose a clinical electromyogram (EMG), without supervision, is proposed. Due to the lack of a priori knowledge of motor unit potential (MUP) morphology, the EMG decomposition must be performed in an unsupervised manner. A neural network classifier, consisting of a multi-layer neural net of perceptrons and using an unsupervised training strategy, is proposed. The neural network learns repetitive appearances of MUP waveforms from their suspected occurrence in a given filtered EMLG signal by using an unsupervised clustering strategy. Upon training, the network creates stable attractors which correspond to nominal representations of MUP clusters hidden in the data. The decomposition/clustering capabilities of the proposed method are validated on a real EMG signal and on an unlabeled signal set.<<ETX>>","PeriodicalId":286849,"journal":{"name":"[Proceedings 1992] IJCNN International Joint Conference on Neural Networks","volume":"125 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132862016","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 : 1992-06-07DOI: 10.1109/IJCNN.1992.226965
M. Franzini
A new connectionist architecture with absolute classification capability is proposed. In the TARGET architecture, each unit has a target vector associated with it, which is the set of output values of units in a lower layer of the network which will cause the unit to be fully activated. When the outputs of all of the sending units closely match a unit's target vector, the unit outputs a value close to zero. The network is trained by gradient descent, using a procedure derived in the same manner as the standard back propagation procedure. A rudimentary test of this system on the exclusive-or-problem is reported, in which a system achieves outputs accurate within 1%. A more extensive test of the system is reported, using a single-speaker isolated-word database of spelled Spanish words, with a vocabulary consisting of the 29 letters of the Spanish alphabet. The recognition rate using the new architecture was 94.0%, compared with 92.5% for standard backpropagation.<>
{"title":"The TARGET architecture: a feature-oriented approach to connectionist word spotting","authors":"M. Franzini","doi":"10.1109/IJCNN.1992.226965","DOIUrl":"https://doi.org/10.1109/IJCNN.1992.226965","url":null,"abstract":"A new connectionist architecture with absolute classification capability is proposed. In the TARGET architecture, each unit has a target vector associated with it, which is the set of output values of units in a lower layer of the network which will cause the unit to be fully activated. When the outputs of all of the sending units closely match a unit's target vector, the unit outputs a value close to zero. The network is trained by gradient descent, using a procedure derived in the same manner as the standard back propagation procedure. A rudimentary test of this system on the exclusive-or-problem is reported, in which a system achieves outputs accurate within 1%. A more extensive test of the system is reported, using a single-speaker isolated-word database of spelled Spanish words, with a vocabulary consisting of the 29 letters of the Spanish alphabet. The recognition rate using the new architecture was 94.0%, compared with 92.5% for standard backpropagation.<<ETX>>","PeriodicalId":286849,"journal":{"name":"[Proceedings 1992] IJCNN International Joint Conference on Neural Networks","volume":"306 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131979674","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 : 1992-06-07DOI: 10.1109/IJCNN.1992.227013
J. Hao, J. Vandewalle
A novel model of discrete neural associative memories is presented. The most important feature of this model is that static mapping instead of the dynamic convergent process is used to retrieve the stored messages. The model features a two-layer structure, with feedforward connections only and using two kinds of neurons. This model uses an extremely simple weight set-up rule and all the resulting weights can only be -1 or +1. Compared to the Hopfield model, the model can guarantee all the given patterns to be stored as fixed points. Each fixed point is surrounded by an attraction ball with the maximum possible radius. The processing speed is much higher because of the use of layered feedforward nets. The model is flexible in the sense that extra patterns can be easily incorporated into the established net.<>
{"title":"A new model of neural associative memories","authors":"J. Hao, J. Vandewalle","doi":"10.1109/IJCNN.1992.227013","DOIUrl":"https://doi.org/10.1109/IJCNN.1992.227013","url":null,"abstract":"A novel model of discrete neural associative memories is presented. The most important feature of this model is that static mapping instead of the dynamic convergent process is used to retrieve the stored messages. The model features a two-layer structure, with feedforward connections only and using two kinds of neurons. This model uses an extremely simple weight set-up rule and all the resulting weights can only be -1 or +1. Compared to the Hopfield model, the model can guarantee all the given patterns to be stored as fixed points. Each fixed point is surrounded by an attraction ball with the maximum possible radius. The processing speed is much higher because of the use of layered feedforward nets. The model is flexible in the sense that extra patterns can be easily incorporated into the established net.<<ETX>>","PeriodicalId":286849,"journal":{"name":"[Proceedings 1992] IJCNN International Joint Conference on Neural Networks","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132033646","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 : 1992-06-07DOI: 10.1109/IJCNN.1992.287077
J.H. Kim, S. Park
A learning algorithm called geometrical expanding learning (GEL) is proposed to train multilayer artificial neural networks (ANNs) with guaranteed convergence for an arbitrary function in a binary field. It is noted that there has not yet been found a learning algorithm for a three-layer ANN which guarantees convergence. The most significant contribution of the proposed research is the development of a learning algorithm for multilayer ANNs which guarantees convergence and automatically determines the required number of neurons. The learning speed of the proposed GEL algorithm is much faster than that of the backpropagation learning algorithm in a binary field.<>
{"title":"The geometrical learning of multi-layer artificial neural networks with guaranteed convergence","authors":"J.H. Kim, S. Park","doi":"10.1109/IJCNN.1992.287077","DOIUrl":"https://doi.org/10.1109/IJCNN.1992.287077","url":null,"abstract":"A learning algorithm called geometrical expanding learning (GEL) is proposed to train multilayer artificial neural networks (ANNs) with guaranteed convergence for an arbitrary function in a binary field. It is noted that there has not yet been found a learning algorithm for a three-layer ANN which guarantees convergence. The most significant contribution of the proposed research is the development of a learning algorithm for multilayer ANNs which guarantees convergence and automatically determines the required number of neurons. The learning speed of the proposed GEL algorithm is much faster than that of the backpropagation learning algorithm in a binary field.<<ETX>>","PeriodicalId":286849,"journal":{"name":"[Proceedings 1992] IJCNN International Joint Conference on Neural Networks","volume":"151 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1992-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132078350","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：