Pub Date : 2002-12-10DOI: 10.1109/NEUREL.2002.1057956
S. Stankovic, M. Milosavljevic, L. Buturovic, M. Stankovic
Summary form only given. This tutorial is devoted to an important segment of statistical learning techniques related to the problem of supervised learning, which aims at predicting the value of an outcome given a number of inputs. Theoretical material is oriented mainly towards methods and concepts. The introduction outlines general aspects of statistical learning, together with motivations for its applications in medicine and genomics. The second part deals with the main theoretical aspects of supervised learning, including a short overview of statistical decision theory, with the emphasis on the problem of trade-off between bias and variance. Attention is further paid to linear methods, applied to both regression and classification problems. In the presentation of neural networks applied to statistical learning, stress is placed on multi-layer perceptrons and training algorithms based on gradient search techniques. Various issues important in practice are given considerable attention, including cross-validation techniques and the choice of suitable learning procedures.
{"title":"Statistical learning: data mining and prediction with applications to medicine and genomics","authors":"S. Stankovic, M. Milosavljevic, L. Buturovic, M. Stankovic","doi":"10.1109/NEUREL.2002.1057956","DOIUrl":"https://doi.org/10.1109/NEUREL.2002.1057956","url":null,"abstract":"Summary form only given. This tutorial is devoted to an important segment of statistical learning techniques related to the problem of supervised learning, which aims at predicting the value of an outcome given a number of inputs. Theoretical material is oriented mainly towards methods and concepts. The introduction outlines general aspects of statistical learning, together with motivations for its applications in medicine and genomics. The second part deals with the main theoretical aspects of supervised learning, including a short overview of statistical decision theory, with the emphasis on the problem of trade-off between bias and variance. Attention is further paid to linear methods, applied to both regression and classification problems. In the presentation of neural networks applied to statistical learning, stress is placed on multi-layer perceptrons and training algorithms based on gradient search techniques. Various issues important in practice are given considerable attention, including cross-validation techniques and the choice of suitable learning procedures.","PeriodicalId":347066,"journal":{"name":"6th Seminar on Neural Network Applications in Electrical Engineering","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121221964","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 : 2002-12-10DOI: 10.1109/NEUREL.2002.1057970
M. Inal, Y.S. Fatihoglu
In this study, self organizing map (SOM) and associative memory model (AMM) artificial neural networks (ANN) are used as hybrid classifier for several speaker recognition experiments. These include text dependent closed-set speaker identification and speaker verification of Turkish speaker set and text independent closed-set speaker identification of a subset of the TIMIT database. Turkish speaker set constitutes 10 speakers with their name and surname. Each utterance is repeated 8 times, 5 of them are used in training and. remaining in the test stages. The subset of the TIMIT database consists 38 speakers from New England region. Each speaker's 10 different utterances are equally selected for using in training and test session. Mel frequency cepstral coefficients (MFCC) method is used for feature extraction of the training and test vectors. When the study is compared with different studies for the same databases, this study gives good results as much as the others.
{"title":"Self organizing map and associative memory model hybrid classifier for speaker recognition","authors":"M. Inal, Y.S. Fatihoglu","doi":"10.1109/NEUREL.2002.1057970","DOIUrl":"https://doi.org/10.1109/NEUREL.2002.1057970","url":null,"abstract":"In this study, self organizing map (SOM) and associative memory model (AMM) artificial neural networks (ANN) are used as hybrid classifier for several speaker recognition experiments. These include text dependent closed-set speaker identification and speaker verification of Turkish speaker set and text independent closed-set speaker identification of a subset of the TIMIT database. Turkish speaker set constitutes 10 speakers with their name and surname. Each utterance is repeated 8 times, 5 of them are used in training and. remaining in the test stages. The subset of the TIMIT database consists 38 speakers from New England region. Each speaker's 10 different utterances are equally selected for using in training and test session. Mel frequency cepstral coefficients (MFCC) method is used for feature extraction of the training and test vectors. When the study is compared with different studies for the same databases, this study gives good results as much as the others.","PeriodicalId":347066,"journal":{"name":"6th Seminar on Neural Network Applications in Electrical Engineering","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128208204","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 : 2002-12-10DOI: 10.1109/NEUREL.2002.1057979
V. Litovski, M. Andrejević
We describe the state of the art and some preliminary results obtained by application of artificial neural networks (ANN) to modelling of dynamic non-linear electronic circuits. ANNs are used for application of the black-box modelling concept in the time domain. The ANN's topology, the testing signal used for excitation, together with the complexity of the ANN is considered. Examples of Pion-linear dynamic modelling are given encompassing a wide variety of modelling problems. Verification of the concept is performed by verifying the ability of the model to generalize i.e. to create acceptable responses to excitations not used during training. Implementation of these models within a behavioural simulator is exemplified.
{"title":"Electronic modelling using ANNs for analogue and mixed-mode behavioural simulation","authors":"V. Litovski, M. Andrejević","doi":"10.1109/NEUREL.2002.1057979","DOIUrl":"https://doi.org/10.1109/NEUREL.2002.1057979","url":null,"abstract":"We describe the state of the art and some preliminary results obtained by application of artificial neural networks (ANN) to modelling of dynamic non-linear electronic circuits. ANNs are used for application of the black-box modelling concept in the time domain. The ANN's topology, the testing signal used for excitation, together with the complexity of the ANN is considered. Examples of Pion-linear dynamic modelling are given encompassing a wide variety of modelling problems. Verification of the concept is performed by verifying the ability of the model to generalize i.e. to create acceptable responses to excitations not used during training. Implementation of these models within a behavioural simulator is exemplified.","PeriodicalId":347066,"journal":{"name":"6th Seminar on Neural Network Applications in Electrical Engineering","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132473653","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 : 2002-12-10DOI: 10.1109/NEUREL.2002.1057957
S. L. Goh, D. Mandic, M. Bozic
The normalized nonlinear gradient descent learning algorithm (NNGD) for a class of nonlinear finite impulse response (FIR) adaptive filters (dynamical perceptron) is extended to the case where the amplitude of the nonlinear activation function is made gradient adaptive. This makes the adaptive amplitude normalized nonlinear gradient descent (AANNGD) algorithm. The AANNGD is suitable for processing of nonlinear and nonstationary signals with a large dynamical range. Experimental results show that AANNGD outperforms the standard LMS, NGD, NNGD, the fully adaptive (FANNGD) and the sign algorithm on nonlinear input with large dynamics.
{"title":"A neural nonlinear adaptive filter with a trainable activation function","authors":"S. L. Goh, D. Mandic, M. Bozic","doi":"10.1109/NEUREL.2002.1057957","DOIUrl":"https://doi.org/10.1109/NEUREL.2002.1057957","url":null,"abstract":"The normalized nonlinear gradient descent learning algorithm (NNGD) for a class of nonlinear finite impulse response (FIR) adaptive filters (dynamical perceptron) is extended to the case where the amplitude of the nonlinear activation function is made gradient adaptive. This makes the adaptive amplitude normalized nonlinear gradient descent (AANNGD) algorithm. The AANNGD is suitable for processing of nonlinear and nonstationary signals with a large dynamical range. Experimental results show that AANNGD outperforms the standard LMS, NGD, NNGD, the fully adaptive (FANNGD) and the sign algorithm on nonlinear input with large dynamics.","PeriodicalId":347066,"journal":{"name":"6th Seminar on Neural Network Applications in Electrical Engineering","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134237215","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 : 2002-12-10DOI: 10.1109/NEUREL.2002.1057960
M. Jankovic
A new approach to unsupervised learning in a single-layer neural. network is discussed. An algorithm for unsupervised learning based on Hebbian learning rule is presented. A simple neuron model is analyzed. Adopted neuron model represents dynamic neural model which contains both feed forward and feedback connections between input and output. Actually, proposed learning algorithm could be more correctly named self-supervised rather than unsupervised. The solution proposed here is a modified Hebbian rule in which the modification of the synaptic strength is proportional not to pre- and post-synaptic activity, but instead to the pre-synaptic and averaged value of post-synaptic activity. It is shown that the model neuron tends to extract the principal component from a stationary input vector sequence. Usually accepted additional decaying terms for the stabilization of original Hebb rule are avoided. Implementation of the basic Hebb scheme would not lead to unrealistic growth of the synaptic strengths, thanks to the adopted network structure.
{"title":"A simple biologically inspired principal component analyzer-ModH neuron model","authors":"M. Jankovic","doi":"10.1109/NEUREL.2002.1057960","DOIUrl":"https://doi.org/10.1109/NEUREL.2002.1057960","url":null,"abstract":"A new approach to unsupervised learning in a single-layer neural. network is discussed. An algorithm for unsupervised learning based on Hebbian learning rule is presented. A simple neuron model is analyzed. Adopted neuron model represents dynamic neural model which contains both feed forward and feedback connections between input and output. Actually, proposed learning algorithm could be more correctly named self-supervised rather than unsupervised. The solution proposed here is a modified Hebbian rule in which the modification of the synaptic strength is proportional not to pre- and post-synaptic activity, but instead to the pre-synaptic and averaged value of post-synaptic activity. It is shown that the model neuron tends to extract the principal component from a stationary input vector sequence. Usually accepted additional decaying terms for the stabilization of original Hebb rule are avoided. Implementation of the basic Hebb scheme would not lead to unrealistic growth of the synaptic strengths, thanks to the adopted network structure.","PeriodicalId":347066,"journal":{"name":"6th Seminar on Neural Network Applications in Electrical Engineering","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131400149","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 : 2002-12-10DOI: 10.1109/NEUREL.2002.1057974
A. Tsakoumis, S. Vladov, V. Mladenov
Multilayer perceptron (MLP) network,and Elman neural network have been compared in electric load forecasting. The electric load profile is considered as a time series and it has been shown that Elman network models load in an electric power utility better than MLP network.
{"title":"Electric load forecasting with multilayer perceptron and Elman neural network","authors":"A. Tsakoumis, S. Vladov, V. Mladenov","doi":"10.1109/NEUREL.2002.1057974","DOIUrl":"https://doi.org/10.1109/NEUREL.2002.1057974","url":null,"abstract":"Multilayer perceptron (MLP) network,and Elman neural network have been compared in electric load forecasting. The electric load profile is considered as a time series and it has been shown that Elman network models load in an electric power utility better than MLP network.","PeriodicalId":347066,"journal":{"name":"6th Seminar on Neural Network Applications in Electrical Engineering","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124343943","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 : 2002-12-10DOI: 10.1109/NEUREL.2002.1057967
N. Jovanovic, V. Milutinovic, Z. Obradovic
The aim of this paper is to introduce a novel reader to the topic of predictive data mining (DM) by discussing technical aspects and requirements of common mining tools. A description of DM scope is followed by comparing DM to related data management and analysis techniques. This is followed by a discussion of a typical predictive DM process, and some of the more successful algorithms and software packages.
{"title":"Foundations of predictive data mining","authors":"N. Jovanovic, V. Milutinovic, Z. Obradovic","doi":"10.1109/NEUREL.2002.1057967","DOIUrl":"https://doi.org/10.1109/NEUREL.2002.1057967","url":null,"abstract":"The aim of this paper is to introduce a novel reader to the topic of predictive data mining (DM) by discussing technical aspects and requirements of common mining tools. A description of DM scope is followed by comparing DM to related data management and analysis techniques. This is followed by a discussion of a typical predictive DM process, and some of the more successful algorithms and software packages.","PeriodicalId":347066,"journal":{"name":"6th Seminar on Neural Network Applications in Electrical Engineering","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115059109","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 : 2002-12-10DOI: 10.1109/NEUREL.2002.1057965
Z. Bojkovic, D. Milovanovic
Multimedia has at its very core the field of signal processing technology. The key attributes of neural processing essential to intelligent multimedia processing are presented. The objective is to show why neural networks are a core technology for efficient representation for audio/visual information. Also, it will be demonstrated how the adaptive neural network technology gives a unified solution to a broad spectrum of multimedia applications including image visualization, subject-based retrieval, face detection and recognitions. Region of interest (ROI) coding as well as multidescription coding functionalities are incorporated, too.
{"title":"Recent trends in neural networks for multimedia processing","authors":"Z. Bojkovic, D. Milovanovic","doi":"10.1109/NEUREL.2002.1057965","DOIUrl":"https://doi.org/10.1109/NEUREL.2002.1057965","url":null,"abstract":"Multimedia has at its very core the field of signal processing technology. The key attributes of neural processing essential to intelligent multimedia processing are presented. The objective is to show why neural networks are a core technology for efficient representation for audio/visual information. Also, it will be demonstrated how the adaptive neural network technology gives a unified solution to a broad spectrum of multimedia applications including image visualization, subject-based retrieval, face detection and recognitions. Region of interest (ROI) coding as well as multidescription coding functionalities are incorporated, too.","PeriodicalId":347066,"journal":{"name":"6th Seminar on Neural Network Applications in Electrical Engineering","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123560716","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 : 2002-12-10DOI: 10.1109/NEUREL.2002.1057971
D. Arsenijević, M. Milosavljevic
2 autoregressive (AR) models of Serbian phonemes are examined in this paper. They are the linear autoregressive model and the nonlinear model realized in a feedforward neural network with one hidden layer. It is shown that both models gave satisfying results.
{"title":"Comparative analysis of Serbian phonemes","authors":"D. Arsenijević, M. Milosavljevic","doi":"10.1109/NEUREL.2002.1057971","DOIUrl":"https://doi.org/10.1109/NEUREL.2002.1057971","url":null,"abstract":"2 autoregressive (AR) models of Serbian phonemes are examined in this paper. They are the linear autoregressive model and the nonlinear model realized in a feedforward neural network with one hidden layer. It is shown that both models gave satisfying results.","PeriodicalId":347066,"journal":{"name":"6th Seminar on Neural Network Applications in Electrical Engineering","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124778471","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 : 2002-12-10DOI: 10.1109/NEUREL.2002.1057962
M.S. Stankovic, B. Todorovic, B.M. Vidojkovic
Time series prediction is based on reconstruction of unknown, possibly chaotic dynamics using a certain number of delayed values of the time series and realizing the mapping between them and future values. The number of previous values used for reconstruction (usually called the embedding dimension) strongly influences the complexity of the mapping. We have applied structurally adaptive RBF networks to determine the embedding dimension and to realize the desired mapping between the past and future values. The method is tested on reconstruction of Henon maps and Lorenz chaotic attractors.
{"title":"Reconstruction of chaotic dynamics using structurally adaptive radial basis function networks","authors":"M.S. Stankovic, B. Todorovic, B.M. Vidojkovic","doi":"10.1109/NEUREL.2002.1057962","DOIUrl":"https://doi.org/10.1109/NEUREL.2002.1057962","url":null,"abstract":"Time series prediction is based on reconstruction of unknown, possibly chaotic dynamics using a certain number of delayed values of the time series and realizing the mapping between them and future values. The number of previous values used for reconstruction (usually called the embedding dimension) strongly influences the complexity of the mapping. We have applied structurally adaptive RBF networks to determine the embedding dimension and to realize the desired mapping between the past and future values. The method is tested on reconstruction of Henon maps and Lorenz chaotic attractors.","PeriodicalId":347066,"journal":{"name":"6th Seminar on Neural Network Applications in Electrical Engineering","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129164041","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：