Pub Date : 2002-11-18DOI: 10.1109/ICONIP.2002.1198982
G. Hernández, P. Munro, J. Rubin
The dependence of synaptic plasticity on both presynaptic activity and postsynaptic activity, as postulated by Hebb, has been clearly and repeatedly demonstrated in the laboratory. Traditionally, "activity" has been measured by counting spikes in a short time window to get an average firing rate. Recent experiments reveal functional synaptic changes that depend on the precise timing of individual pairs of spikes (one presynaptic and one postsynaptic). Here, the emergence of rate-based learning rules from spike-based dependencies is introduced, through the idea of a rate map in synaptic weight space.
{"title":"Mapping from the spike domain to the rate-based domain","authors":"G. Hernández, P. Munro, J. Rubin","doi":"10.1109/ICONIP.2002.1198982","DOIUrl":"https://doi.org/10.1109/ICONIP.2002.1198982","url":null,"abstract":"The dependence of synaptic plasticity on both presynaptic activity and postsynaptic activity, as postulated by Hebb, has been clearly and repeatedly demonstrated in the laboratory. Traditionally, \"activity\" has been measured by counting spikes in a short time window to get an average firing rate. Recent experiments reveal functional synaptic changes that depend on the precise timing of individual pairs of spikes (one presynaptic and one postsynaptic). Here, the emergence of rate-based learning rules from spike-based dependencies is introduced, through the idea of a rate map in synaptic weight space.","PeriodicalId":146553,"journal":{"name":"Proceedings of the 9th International Conference on Neural Information Processing, 2002. ICONIP '02.","volume":"25 4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122616089","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 : 2002-11-18DOI: 10.1109/ICONIP.2002.1202844
E. Oja
The FastICA algorithm is one of the most popular methods to solve problems in independent component analysis (ICA) and blind source separation. It has been shown experimentally that it outperforms most of the commonly used ICA algorithms in convergence speed. A rigorous convergence analysis has been presented only for the so-called one-unit case, in which just one of the rows of the separating matrix is considered. However, in the FastICA algorithm, there is also an explicit normalization step, and it may be questioned whether the extra rotation caused by the normalization will effect the convergence speed. The purpose of this paper is to show that this is not the case and the good convergence properties of the one-unit case are also shared by the full algorithm with symmetrical normalization.
{"title":"Convergence of the symmetrical FastICA algorithm","authors":"E. Oja","doi":"10.1109/ICONIP.2002.1202844","DOIUrl":"https://doi.org/10.1109/ICONIP.2002.1202844","url":null,"abstract":"The FastICA algorithm is one of the most popular methods to solve problems in independent component analysis (ICA) and blind source separation. It has been shown experimentally that it outperforms most of the commonly used ICA algorithms in convergence speed. A rigorous convergence analysis has been presented only for the so-called one-unit case, in which just one of the rows of the separating matrix is considered. However, in the FastICA algorithm, there is also an explicit normalization step, and it may be questioned whether the extra rotation caused by the normalization will effect the convergence speed. The purpose of this paper is to show that this is not the case and the good convergence properties of the one-unit case are also shared by the full algorithm with symmetrical normalization.","PeriodicalId":146553,"journal":{"name":"Proceedings of the 9th International Conference on Neural Information Processing, 2002. ICONIP '02.","volume":"28 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131121839","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 : 2002-11-18DOI: 10.1109/ICONIP.2002.1198155
H. Nakagama, S. Tanaka
We study the performance of three types of recurrent neural networks (RNN) for the production of natural language sentences: Simple Recurrent Networks (SRN), Back-Propagation Through Time (BPTT) and Sequential Recursive Auto-Associative Memory (SRAAM). We used simple and complex grammars to compare the ability of learning and being scaled up. Among them, SRAAM is found to have highest performance of training and producing fairly complex and long sentences.
{"title":"An experimental comparison of recurrent neural network for natural language production","authors":"H. Nakagama, S. Tanaka","doi":"10.1109/ICONIP.2002.1198155","DOIUrl":"https://doi.org/10.1109/ICONIP.2002.1198155","url":null,"abstract":"We study the performance of three types of recurrent neural networks (RNN) for the production of natural language sentences: Simple Recurrent Networks (SRN), Back-Propagation Through Time (BPTT) and Sequential Recursive Auto-Associative Memory (SRAAM). We used simple and complex grammars to compare the ability of learning and being scaled up. Among them, SRAAM is found to have highest performance of training and producing fairly complex and long sentences.","PeriodicalId":146553,"journal":{"name":"Proceedings of the 9th International Conference on Neural Information Processing, 2002. ICONIP '02.","volume":"43 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131460286","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 : 2002-11-18DOI: 10.1109/ICONIP.2002.1202122
B. Roerig, B. Chen, J. Kao
Simple cells in layer 4 of the primary visual cortex are the first neurons in the visual pathway showing orientation and direction selective responses. The precise role of intracortical excitatory and inhibitory connections in generating these properties is still unclear. Intracortical inhibitory processes have been shown to be crucial to the generation of direction selective responses. In vivo, excitatory and inhibitory layer 4 cells differ in their receptive field properties: excitatory (regular spiking) neurons are orientation- and direction selective whereas inhibitory (fast spiking) neurons are orientation-, but poorly direction tuned. This difference in direction tuning could be due to differences in intracortical inhibitory synaptic input patterns. To address this question we have optically recorded orientation and direction maps from ferret primary visual cortex. Subsequently the imaged brain region was removed and tangential slices prepared. Whole cell patch clamp recordings from individual layer 4 neurons were done and synaptic inputs were scanned by local photolysis of caged glutamate. Postsynaptic cells were filled with biocytin and histological sections were aligned with the synaptic input maps and the optical images obtained in vivo to determine the spatial distribution of presynaptic inputs. The majority (68%) of excitatory inputs to both spiny (excitatory) and aspiny (inhibitory) stellate cells originated from cortical regions preferring the same orientation and direction as the postsynaptic cell. However, the inhibitory input patterns were significantly different for the two cell populations: excitatory layer 4 cells received two populations of inhibitory inputs, about 50% originated in iso-direction domains whereas the remaining inputs originated in cortical regions preferring the opposite direction of stimulus motion. This indicates that specific inhibitory connections originating in regions tuned to the opposite direction are important for direction tuning of cortical neurons and that differences in response properties in different populations of cortical neurons might be explained by their different intracortical connectivity patterns.
{"title":"Organization of inhibitory synaptic circuits in layer 4 of ferret visual cortex related to direction preference maps","authors":"B. Roerig, B. Chen, J. Kao","doi":"10.1109/ICONIP.2002.1202122","DOIUrl":"https://doi.org/10.1109/ICONIP.2002.1202122","url":null,"abstract":"Simple cells in layer 4 of the primary visual cortex are the first neurons in the visual pathway showing orientation and direction selective responses. The precise role of intracortical excitatory and inhibitory connections in generating these properties is still unclear. Intracortical inhibitory processes have been shown to be crucial to the generation of direction selective responses. In vivo, excitatory and inhibitory layer 4 cells differ in their receptive field properties: excitatory (regular spiking) neurons are orientation- and direction selective whereas inhibitory (fast spiking) neurons are orientation-, but poorly direction tuned. This difference in direction tuning could be due to differences in intracortical inhibitory synaptic input patterns. To address this question we have optically recorded orientation and direction maps from ferret primary visual cortex. Subsequently the imaged brain region was removed and tangential slices prepared. Whole cell patch clamp recordings from individual layer 4 neurons were done and synaptic inputs were scanned by local photolysis of caged glutamate. Postsynaptic cells were filled with biocytin and histological sections were aligned with the synaptic input maps and the optical images obtained in vivo to determine the spatial distribution of presynaptic inputs. The majority (68%) of excitatory inputs to both spiny (excitatory) and aspiny (inhibitory) stellate cells originated from cortical regions preferring the same orientation and direction as the postsynaptic cell. However, the inhibitory input patterns were significantly different for the two cell populations: excitatory layer 4 cells received two populations of inhibitory inputs, about 50% originated in iso-direction domains whereas the remaining inputs originated in cortical regions preferring the opposite direction of stimulus motion. This indicates that specific inhibitory connections originating in regions tuned to the opposite direction are important for direction tuning of cortical neurons and that differences in response properties in different populations of cortical neurons might be explained by their different intracortical connectivity patterns.","PeriodicalId":146553,"journal":{"name":"Proceedings of the 9th International Conference on Neural Information Processing, 2002. ICONIP '02.","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131492138","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 : 2002-11-18DOI: 10.1109/ICONIP.2002.1199013
A. Reddipogu, G. Maxwell, C. MacLeod, M. Simpson
This paper discusses the development of a novel pattern recognition system using artificial neural networks (ANNs) and evolutionary algorithms for reinforcement learning (EARL). The network is based on neuronal interactions involved in identification of prey and predator in toads. The distributed neural network (DNN) is capable of recognizing and classifying various features. The lateral inhibition between the output neurons helps the network in the classification process - similar to the gate in gating network. The results obtained are compared with standard neural network architectures and learning algorithms.
{"title":"A novel artificial neural network trained using evolutionary algorithms for reinforcement learning","authors":"A. Reddipogu, G. Maxwell, C. MacLeod, M. Simpson","doi":"10.1109/ICONIP.2002.1199013","DOIUrl":"https://doi.org/10.1109/ICONIP.2002.1199013","url":null,"abstract":"This paper discusses the development of a novel pattern recognition system using artificial neural networks (ANNs) and evolutionary algorithms for reinforcement learning (EARL). The network is based on neuronal interactions involved in identification of prey and predator in toads. The distributed neural network (DNN) is capable of recognizing and classifying various features. The lateral inhibition between the output neurons helps the network in the classification process - similar to the gate in gating network. The results obtained are compared with standard neural network architectures and learning algorithms.","PeriodicalId":146553,"journal":{"name":"Proceedings of the 9th International Conference on Neural Information Processing, 2002. ICONIP '02.","volume":"23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127593289","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 : 2002-11-18DOI: 10.1109/ICONIP.2002.1202190
T. Horiguchi, H. Yokoyama
We propose a four-layered neural network model and a five-layered neural network model for the hippocampal system by extending the three-layered model given by Araki and Aihara (1998), in order to introduce an effect of auto-recurrent connections in CA3 and also an effect of cholinergic modulation in CA1 and CA3 from the medial septum. We investigate the storage and the recall of dynamical patterns for the proposed models with or without inhibitory connections in CA3. We clarify two different sequential patterns that the storage and the recall succeeded for the proposed models, but not for the original three-layered model. We discuss an effect of acetylcholine to neurons in CA1 and CA3 transmitted from the medial septum.
{"title":"Storage and recall of dynamical patterns in neural network models of hippocampus","authors":"T. Horiguchi, H. Yokoyama","doi":"10.1109/ICONIP.2002.1202190","DOIUrl":"https://doi.org/10.1109/ICONIP.2002.1202190","url":null,"abstract":"We propose a four-layered neural network model and a five-layered neural network model for the hippocampal system by extending the three-layered model given by Araki and Aihara (1998), in order to introduce an effect of auto-recurrent connections in CA3 and also an effect of cholinergic modulation in CA1 and CA3 from the medial septum. We investigate the storage and the recall of dynamical patterns for the proposed models with or without inhibitory connections in CA3. We clarify two different sequential patterns that the storage and the recall succeeded for the proposed models, but not for the original three-layered model. We discuss an effect of acetylcholine to neurons in CA1 and CA3 transmitted from the medial septum.","PeriodicalId":146553,"journal":{"name":"Proceedings of the 9th International Conference on Neural Information Processing, 2002. ICONIP '02.","volume":"248 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133515394","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 : 2002-11-18DOI: 10.1109/ICONIP.2002.1202168
A. Frolov, D. Rachkovskij, D. Húsek
A sparsely encoded Willshaw-like attractor neural network based on binary Hebbian synapses is investigated analytically and by computer simulations. A special inhibition mechanism which supports a constant number of active neurons at each time step is used. Informational capacity and size of attraction basins are evaluated for the single-step and the Gibson-Robinson approximations, as well as for experimental results.
{"title":"On information characteristics of sparsely encoded binary auto-associative memory","authors":"A. Frolov, D. Rachkovskij, D. Húsek","doi":"10.1109/ICONIP.2002.1202168","DOIUrl":"https://doi.org/10.1109/ICONIP.2002.1202168","url":null,"abstract":"A sparsely encoded Willshaw-like attractor neural network based on binary Hebbian synapses is investigated analytically and by computer simulations. A special inhibition mechanism which supports a constant number of active neurons at each time step is used. Informational capacity and size of attraction basins are evaluated for the single-step and the Gibson-Robinson approximations, as well as for experimental results.","PeriodicalId":146553,"journal":{"name":"Proceedings of the 9th International Conference on Neural Information Processing, 2002. ICONIP '02.","volume":"66 15","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133587819","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 : 2002-11-18DOI: 10.1109/ICONIP.2002.1198184
P. Patil, U. Kulkarni, T. Sontakke
Fuzzy mean point clustering neural network (FMPCNN) is proposed with its learning algorithm, which utilizes fuzzy sets as pattern clusters. The performance of FMPCNN when verified with Fisher Iris data, it is found superior to Simpson's fuzzy min-max neural network and fuzzy hyperline segment clustering neural network (FHLSCNN) proposed by Kulkarni and Sontakke.
{"title":"Fuzzy mean point clustering neural network","authors":"P. Patil, U. Kulkarni, T. Sontakke","doi":"10.1109/ICONIP.2002.1198184","DOIUrl":"https://doi.org/10.1109/ICONIP.2002.1198184","url":null,"abstract":"Fuzzy mean point clustering neural network (FMPCNN) is proposed with its learning algorithm, which utilizes fuzzy sets as pattern clusters. The performance of FMPCNN when verified with Fisher Iris data, it is found superior to Simpson's fuzzy min-max neural network and fuzzy hyperline segment clustering neural network (FHLSCNN) proposed by Kulkarni and Sontakke.","PeriodicalId":146553,"journal":{"name":"Proceedings of the 9th International Conference on Neural Information Processing, 2002. ICONIP '02.","volume":"23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133763778","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 : 2002-11-18DOI: 10.1109/ICONIP.2002.1202154
T. Tanprasert, T. Kripruksawan
The paper presents a decayed prior sampling algorithm for integrating the existing knowledge of a supervised learning neural networks with the new training data. The algorithm allows the existing knowledge to age out in slow rate as a neural network is gradually retrained with consecutive sets of new samples, resembling the change of application locality under a consistent environment. The experiments are performed on 2-dimensional partitions problem and the results convincingly confirm the effectiveness of the technique.
{"title":"An approach to control aging rate of neural networks under adaptation to gradually changing context","authors":"T. Tanprasert, T. Kripruksawan","doi":"10.1109/ICONIP.2002.1202154","DOIUrl":"https://doi.org/10.1109/ICONIP.2002.1202154","url":null,"abstract":"The paper presents a decayed prior sampling algorithm for integrating the existing knowledge of a supervised learning neural networks with the new training data. The algorithm allows the existing knowledge to age out in slow rate as a neural network is gradually retrained with consecutive sets of new samples, resembling the change of application locality under a consistent environment. The experiments are performed on 2-dimensional partitions problem and the results convincingly confirm the effectiveness of the technique.","PeriodicalId":146553,"journal":{"name":"Proceedings of the 9th International Conference on Neural Information Processing, 2002. ICONIP '02.","volume":"70 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133903943","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 : 2002-11-18DOI: 10.1109/ICONIP.2002.1198126
N. Kasabov
The paper describes what evolving processes are and presents a computational model called evolving connectionist systems (ECOS). The model is based on principles from both brain organization and genetics. The applicability of the model for dynamic modeling and knowledge discovery in the areas of brain study, bioinformatics, speech and language learning, adaptive control and adaptive decision support is discussed.
{"title":"Evolving connectionist systems for adaptive learning and knowledge discovery: methods, tools, applications","authors":"N. Kasabov","doi":"10.1109/ICONIP.2002.1198126","DOIUrl":"https://doi.org/10.1109/ICONIP.2002.1198126","url":null,"abstract":"The paper describes what evolving processes are and presents a computational model called evolving connectionist systems (ECOS). The model is based on principles from both brain organization and genetics. The applicability of the model for dynamic modeling and knowledge discovery in the areas of brain study, bioinformatics, speech and language learning, adaptive control and adaptive decision support is discussed.","PeriodicalId":146553,"journal":{"name":"Proceedings of the 9th International Conference on Neural Information Processing, 2002. ICONIP '02.","volume":"190 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133971745","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：