Pub Date : 1999-11-16DOI: 10.1109/ICONIP.1999.844671
Liming Zhang, WeiPing Fan
A new model, called the space-partitioning multilayer perceptron (SP-MLP), is proposed in this paper to resolve classification problems. The number of first-hidden-layer units is determined adaptively, and we introduce a new sub-algorithm to improve the robustness of the network. The results of experiments show that the SP-MLP is more robust than other models. The issue of generalization is also discussed in this paper.
{"title":"A robust model based on space-partitioning method","authors":"Liming Zhang, WeiPing Fan","doi":"10.1109/ICONIP.1999.844671","DOIUrl":"https://doi.org/10.1109/ICONIP.1999.844671","url":null,"abstract":"A new model, called the space-partitioning multilayer perceptron (SP-MLP), is proposed in this paper to resolve classification problems. The number of first-hidden-layer units is determined adaptively, and we introduce a new sub-algorithm to improve the robustness of the network. The results of experiments show that the SP-MLP is more robust than other models. The issue of generalization is also discussed in this paper.","PeriodicalId":237855,"journal":{"name":"ICONIP'99. ANZIIS'99 & ANNES'99 & ACNN'99. 6th International Conference on Neural Information Processing. Proceedings (Cat. No.99EX378)","volume":"49 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1999-11-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129277649","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 : 1999-11-16DOI: 10.1109/ICONIP.1999.845686
Samuel Kaski, Jarkko Venna, T. Kohonen
Introduces a method for assigning colors to displays of cluster structures of high-dimensional data, such that the perceptual differences of the colors reflect the distances in the original data space as faithfully as possible. The cluster structure is first discovered with a self-organizing map (SOM), and then a new nonlinear projection method is applied to map the cluster structure into the CIELab color space. The projection method preserves best the local data distances that are the most important ones, while the global order is still discernible from the colors, too. This allows the method to conform flexibly to the available color space. The output space of the projection need not necessarily be the color space, however. Projections onto, say, two dimensions can be visualized as well.
{"title":"Coloring that reveals high-dimensional structures in data","authors":"Samuel Kaski, Jarkko Venna, T. Kohonen","doi":"10.1109/ICONIP.1999.845686","DOIUrl":"https://doi.org/10.1109/ICONIP.1999.845686","url":null,"abstract":"Introduces a method for assigning colors to displays of cluster structures of high-dimensional data, such that the perceptual differences of the colors reflect the distances in the original data space as faithfully as possible. The cluster structure is first discovered with a self-organizing map (SOM), and then a new nonlinear projection method is applied to map the cluster structure into the CIELab color space. The projection method preserves best the local data distances that are the most important ones, while the global order is still discernible from the colors, too. This allows the method to conform flexibly to the available color space. The output space of the projection need not necessarily be the color space, however. Projections onto, say, two dimensions can be visualized as well.","PeriodicalId":237855,"journal":{"name":"ICONIP'99. ANZIIS'99 & ANNES'99 & ACNN'99. 6th International Conference on Neural Information Processing. Proceedings (Cat. No.99EX378)","volume":"39 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1999-11-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129657497","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 : 1999-11-16DOI: 10.1109/ICONIP.1999.844684
F. Frost, V. Karri
There is an increasing need to apply emerging technologies to achieve process improvements in a dynamic industrial environment. In particular, process control is increasingly popular as an area of manufacturing that can be significantly enhanced using neural networks. Neural networks offer a technology that has the capability, in the first instance, to model process behaviour without a-priori knowledge of the process or the need for complex calculations to model the process mathematically. This paper focuses on two particular networks in particular: backpropagation (BP) and general regression neural network (GRNN) models. As a measure of the performance of these two models, prediction accuracy is evaluated using a practical application in the aluminium smelting industry. The dynamic behaviour of aluminium smelting makes the particular application well-suited to neural network modelling.
{"title":"Performance comparison of BP and GRNN models of the neural network paradigm using a practical industrial application","authors":"F. Frost, V. Karri","doi":"10.1109/ICONIP.1999.844684","DOIUrl":"https://doi.org/10.1109/ICONIP.1999.844684","url":null,"abstract":"There is an increasing need to apply emerging technologies to achieve process improvements in a dynamic industrial environment. In particular, process control is increasingly popular as an area of manufacturing that can be significantly enhanced using neural networks. Neural networks offer a technology that has the capability, in the first instance, to model process behaviour without a-priori knowledge of the process or the need for complex calculations to model the process mathematically. This paper focuses on two particular networks in particular: backpropagation (BP) and general regression neural network (GRNN) models. As a measure of the performance of these two models, prediction accuracy is evaluated using a practical application in the aluminium smelting industry. The dynamic behaviour of aluminium smelting makes the particular application well-suited to neural network modelling.","PeriodicalId":237855,"journal":{"name":"ICONIP'99. ANZIIS'99 & ANNES'99 & ACNN'99. 6th International Conference on Neural Information Processing. Proceedings (Cat. No.99EX378)","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1999-11-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121017546","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 : 1999-11-16DOI: 10.1109/ICONIP.1999.844006
S. Vijayakumar, Si Wu
We provide an RKHS based inverse problem formulation for analytically deriving the optimal function approximation when probabilistic information about the underlying regression is available in terms of the associated correlation functions as used by Poggio and Girosi (1998) and Peney and Atick (1996). On the lines of Poggio and Girosi, we show that this solution can be sparsified using principles of SVM and provide an implementation of this sparsification using a novel, conceptually simple and robust gradient based sequential method instead of the conventional quadratic programming routines.
{"title":"A gradient based technique for generating sparse representation in function approximation","authors":"S. Vijayakumar, Si Wu","doi":"10.1109/ICONIP.1999.844006","DOIUrl":"https://doi.org/10.1109/ICONIP.1999.844006","url":null,"abstract":"We provide an RKHS based inverse problem formulation for analytically deriving the optimal function approximation when probabilistic information about the underlying regression is available in terms of the associated correlation functions as used by Poggio and Girosi (1998) and Peney and Atick (1996). On the lines of Poggio and Girosi, we show that this solution can be sparsified using principles of SVM and provide an implementation of this sparsification using a novel, conceptually simple and robust gradient based sequential method instead of the conventional quadratic programming routines.","PeriodicalId":237855,"journal":{"name":"ICONIP'99. ANZIIS'99 & ANNES'99 & ACNN'99. 6th International Conference on Neural Information Processing. Proceedings (Cat. No.99EX378)","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1999-11-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121338824","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 : 1999-11-16DOI: 10.1109/ICONIP.1999.844665
Zili Zhang, Chengqi Zhang
Many papers have been published on soft computing and software agents respectively, but few involved in how to incorporate soft computing into software agents in practice. The approaches to incorporating soft computing technologies into individual software agents as well as multiagent systems are presented. The benefits and limitations of each approach are also discussed. We tested the multiagent model using JATLite.
{"title":"Approaches to incorporating soft computing technologies into software agents","authors":"Zili Zhang, Chengqi Zhang","doi":"10.1109/ICONIP.1999.844665","DOIUrl":"https://doi.org/10.1109/ICONIP.1999.844665","url":null,"abstract":"Many papers have been published on soft computing and software agents respectively, but few involved in how to incorporate soft computing into software agents in practice. The approaches to incorporating soft computing technologies into individual software agents as well as multiagent systems are presented. The benefits and limitations of each approach are also discussed. We tested the multiagent model using JATLite.","PeriodicalId":237855,"journal":{"name":"ICONIP'99. ANZIIS'99 & ANNES'99 & ACNN'99. 6th International Conference on Neural Information Processing. Proceedings (Cat. No.99EX378)","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1999-11-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121657813","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 : 1999-11-16DOI: 10.1109/ICONIP.1999.845689
N. Pal, A. Laha
Proposes a new scheme for designing a nearest-prototype classifier. The system starts with the minimum number of prototypes, equal to the number of classes. Kohonen's self-organizing feature map (SOFM) algorithm is used to obtain this initial set of prototypes. Then, on the basis of the classification performance, new prototypes are generated dynamically, similar prototypes are merged, and prototypes with less significance are deleted, leading to better performance. If prototypes are deleted or new prototypes appear, then they are retrained using Kohonen's SOFM algorithm with the winner-only update scheme. This adaptation continues until the system satisfies a termination condition. The classifier has been tested with several well-known data sets. The results obtained are quite satisfactory.
{"title":"Design of a nearest-prototype classifier with dynamically generated prototypes using self-organizing feature maps","authors":"N. Pal, A. Laha","doi":"10.1109/ICONIP.1999.845689","DOIUrl":"https://doi.org/10.1109/ICONIP.1999.845689","url":null,"abstract":"Proposes a new scheme for designing a nearest-prototype classifier. The system starts with the minimum number of prototypes, equal to the number of classes. Kohonen's self-organizing feature map (SOFM) algorithm is used to obtain this initial set of prototypes. Then, on the basis of the classification performance, new prototypes are generated dynamically, similar prototypes are merged, and prototypes with less significance are deleted, leading to better performance. If prototypes are deleted or new prototypes appear, then they are retrained using Kohonen's SOFM algorithm with the winner-only update scheme. This adaptation continues until the system satisfies a termination condition. The classifier has been tested with several well-known data sets. The results obtained are quite satisfactory.","PeriodicalId":237855,"journal":{"name":"ICONIP'99. ANZIIS'99 & ANNES'99 & ACNN'99. 6th International Conference on Neural Information Processing. Proceedings (Cat. No.99EX378)","volume":"26 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1999-11-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122578957","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 : 1999-11-16DOI: 10.1109/ICONIP.1999.845694
H. Itoh, K. Aihara
We combine an actor/critic algorithm with goal-directed reasoning. There is a claim that the actor/critic algorithm can be a model of the basal ganglia. The basal ganglia seems to contribute to the higher functions such as goal-directed reasoning. Therefore, an important problem is understanding goal-directed reasoning in the framework of the actor/critic algorithm. As the goal-directed reasoning is realized by changing the current goal, we consider changing the goal as an action and incorporate it into the actor/critic algorithm. One fundamental algorithm and its extensions are proposed with simulation results.
{"title":"Combination of actor/critic algorithm with the goal-directed reasoning","authors":"H. Itoh, K. Aihara","doi":"10.1109/ICONIP.1999.845694","DOIUrl":"https://doi.org/10.1109/ICONIP.1999.845694","url":null,"abstract":"We combine an actor/critic algorithm with goal-directed reasoning. There is a claim that the actor/critic algorithm can be a model of the basal ganglia. The basal ganglia seems to contribute to the higher functions such as goal-directed reasoning. Therefore, an important problem is understanding goal-directed reasoning in the framework of the actor/critic algorithm. As the goal-directed reasoning is realized by changing the current goal, we consider changing the goal as an action and incorporate it into the actor/critic algorithm. One fundamental algorithm and its extensions are proposed with simulation results.","PeriodicalId":237855,"journal":{"name":"ICONIP'99. ANZIIS'99 & ANNES'99 & ACNN'99. 6th International Conference on Neural Information Processing. Proceedings (Cat. No.99EX378)","volume":"77 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1999-11-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122807463","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 : 1999-11-16DOI: 10.1109/ICONIP.1999.844000
J. Kropp
Kohonen's self-organising map (SOM), combined with a measure of topological ordering, is applied to solve a complex classification problem. Climate classifications are mostly empirically-based and often mix the mutual impact between climate, soil and vegetation. Therefore, the influence of abiotic factors on the broad-scale vegetation distribution is of major interest. In order to assess this problem, a spatially highly-resolved climate and soil database is used as training data for a SOM. Inherent feature types hidden in the database are identified, leading to a global pattern of archetypal climatic and soil domains. Additionally, such a classification scheme can be used for comparison with vegetation models and allows a network-based estimation of the potential broad-scale distribution of ecosystem complexes.
{"title":"A neural solution: a data driven assessment of global climate and vegetation classes","authors":"J. Kropp","doi":"10.1109/ICONIP.1999.844000","DOIUrl":"https://doi.org/10.1109/ICONIP.1999.844000","url":null,"abstract":"Kohonen's self-organising map (SOM), combined with a measure of topological ordering, is applied to solve a complex classification problem. Climate classifications are mostly empirically-based and often mix the mutual impact between climate, soil and vegetation. Therefore, the influence of abiotic factors on the broad-scale vegetation distribution is of major interest. In order to assess this problem, a spatially highly-resolved climate and soil database is used as training data for a SOM. Inherent feature types hidden in the database are identified, leading to a global pattern of archetypal climatic and soil domains. Additionally, such a classification scheme can be used for comparison with vegetation models and allows a network-based estimation of the potential broad-scale distribution of ecosystem complexes.","PeriodicalId":237855,"journal":{"name":"ICONIP'99. ANZIIS'99 & ANNES'99 & ACNN'99. 6th International Conference on Neural Information Processing. Proceedings (Cat. No.99EX378)","volume":"54 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1999-11-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124238083","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 : 1999-11-16DOI: 10.1109/ICONIP.1999.845648
M. Tagscherer, L. Kindermann, A. Lewandowski, P. Protzel
In this paper we present an incremental construction algorithm for continuous learning tasks and one of its special features-simultaneous learning of the target function and a confidence value for the system predictions. The basis of the hybrid system is a radial basis function (RBF) network layer. The second layer consists of local models. The two layers are closely combined with a strong interaction. The number of RBF-neurons and the number of local models have not to be determined in advance. This is one of the main advantages of the algorithm. Another advantage emphasized in this paper is the ability to learn the training data distribution simultaneously to the learning of the target function. The learned data set distribution can be used as a confidence value for a given network prediction. The development of the described approach is embedded in a larger project that is primarily concerned with system identification tasks for industrial control such as steel processing.
{"title":"Overcome neural limitations for real world applications by providing confidence values for network prediction","authors":"M. Tagscherer, L. Kindermann, A. Lewandowski, P. Protzel","doi":"10.1109/ICONIP.1999.845648","DOIUrl":"https://doi.org/10.1109/ICONIP.1999.845648","url":null,"abstract":"In this paper we present an incremental construction algorithm for continuous learning tasks and one of its special features-simultaneous learning of the target function and a confidence value for the system predictions. The basis of the hybrid system is a radial basis function (RBF) network layer. The second layer consists of local models. The two layers are closely combined with a strong interaction. The number of RBF-neurons and the number of local models have not to be determined in advance. This is one of the main advantages of the algorithm. Another advantage emphasized in this paper is the ability to learn the training data distribution simultaneously to the learning of the target function. The learned data set distribution can be used as a confidence value for a given network prediction. The development of the described approach is embedded in a larger project that is primarily concerned with system identification tasks for industrial control such as steel processing.","PeriodicalId":237855,"journal":{"name":"ICONIP'99. ANZIIS'99 & ANNES'99 & ACNN'99. 6th International Conference on Neural Information Processing. Proceedings (Cat. No.99EX378)","volume":"159 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1999-11-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123021873","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 : 1999-11-16DOI: 10.1109/ICONIP.1999.845655
A. Bermak, D. Martinez
A systolic multi-precision digital VLSI classifier referred to as "SysNeuro" is presented. Unlike the usual VLSI implementation of classifiers, this hardware has been designed to achieve variable precision computations. A hardware reconfiguration is obtained by using switch elements to change the hardware connection between adjacent 4 bit neuron building blocks. With this reconfiguration concept it is possible to either increase the precision by pooling together adjacent cells or to increase the number of neurons for low levels of precision. Moreover, the design is easily programmable and can be configured to any artificial neural network (ANN) topology in order to cover various kinds of application. The chip integrates 16/8/4 neurons with a corresponding precision of 4/8/16 bits. A prototype has been successfully realized using 0.7 /spl mu/m CMOS technology.
{"title":"Digital VLSI implementation of a multi-precision neural network classifier","authors":"A. Bermak, D. Martinez","doi":"10.1109/ICONIP.1999.845655","DOIUrl":"https://doi.org/10.1109/ICONIP.1999.845655","url":null,"abstract":"A systolic multi-precision digital VLSI classifier referred to as \"SysNeuro\" is presented. Unlike the usual VLSI implementation of classifiers, this hardware has been designed to achieve variable precision computations. A hardware reconfiguration is obtained by using switch elements to change the hardware connection between adjacent 4 bit neuron building blocks. With this reconfiguration concept it is possible to either increase the precision by pooling together adjacent cells or to increase the number of neurons for low levels of precision. Moreover, the design is easily programmable and can be configured to any artificial neural network (ANN) topology in order to cover various kinds of application. The chip integrates 16/8/4 neurons with a corresponding precision of 4/8/16 bits. A prototype has been successfully realized using 0.7 /spl mu/m CMOS technology.","PeriodicalId":237855,"journal":{"name":"ICONIP'99. ANZIIS'99 & ANNES'99 & ACNN'99. 6th International Conference on Neural Information Processing. Proceedings (Cat. No.99EX378)","volume":"177 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1999-11-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115695857","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：