Pub Date : 2003-09-17DOI: 10.1109/NNSP.2003.1318022
Michael E. Tipping, Neil D. Lawrence
We detail a Bayesian interpolation procedure for linear-in-the-parameter models, which combines both effective complexity control and robustness to outliers. Robustness is obtained by adopting a student-t noise distribution, defined hierarchically in terms of an inverse-gamma prior distribution over individual Gaussian observation variances. Importantly, this hierarchical definition enables practical Bayesian variational techniques to concurrently determine both the primary model parameters and the form of the noise process. We show that the model is capable of flexibly inferring, from limited data, both Gaussian and more heavily-tailed student-t noise processes as appropriate.
{"title":"A variational approach to robust Bayesian interpolation","authors":"Michael E. Tipping, Neil D. Lawrence","doi":"10.1109/NNSP.2003.1318022","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318022","url":null,"abstract":"We detail a Bayesian interpolation procedure for linear-in-the-parameter models, which combines both effective complexity control and robustness to outliers. Robustness is obtained by adopting a student-t noise distribution, defined hierarchically in terms of an inverse-gamma prior distribution over individual Gaussian observation variances. Importantly, this hierarchical definition enables practical Bayesian variational techniques to concurrently determine both the primary model parameters and the form of the noise process. We show that the model is capable of flexibly inferring, from limited data, both Gaussian and more heavily-tailed student-t noise processes as appropriate.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"283 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116230546","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 : 2003-09-17DOI: 10.1109/NNSP.2003.1318018
M. Lázaro, I. Santamaría, F. Pérez-Cruz, Antonio Artés-Rodríguez
In this paper, the problem of simultaneously approximating a function and its derivative is formulated within the support vector machine (SVM) framework. The problem has been solved by using the /spl epsiv/-insensitive loss function and introducing new linear constraints in the approximation of the derivative. The resulting quadratic problem can be solved by quadratic programming (QP) techniques. Moreover, a computationally efficient iterative re-weighted least square (IRWLS) procedure has been derived to solve the problem in large data sets. The performance of the method has been compared with the conventional SVM for regression, providing outstanding results.
{"title":"Support vector machine for the simultaneous approximation of a function and its derivative","authors":"M. Lázaro, I. Santamaría, F. Pérez-Cruz, Antonio Artés-Rodríguez","doi":"10.1109/NNSP.2003.1318018","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318018","url":null,"abstract":"In this paper, the problem of simultaneously approximating a function and its derivative is formulated within the support vector machine (SVM) framework. The problem has been solved by using the /spl epsiv/-insensitive loss function and introducing new linear constraints in the approximation of the derivative. The resulting quadratic problem can be solved by quadratic programming (QP) techniques. Moreover, a computationally efficient iterative re-weighted least square (IRWLS) procedure has been derived to solve the problem in large data sets. The performance of the method has been compared with the conventional SVM for regression, providing outstanding results.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132137622","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 : 2003-09-17DOI: 10.1109/NNSP.2003.1318044
J. Kohlmorgen
We present an algorithm that efficiently computes optimal partitions of sequential data into 1 to N segments and propose a method to determine the most salient segmentation among them. As a by-product, we obtain a regularization parameter that can be used to compute such salient segmentations - also on new data sets - even more efficiently.
{"title":"On optimal segmentation of sequential data","authors":"J. Kohlmorgen","doi":"10.1109/NNSP.2003.1318044","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318044","url":null,"abstract":"We present an algorithm that efficiently computes optimal partitions of sequential data into 1 to N segments and propose a method to determine the most salient segmentation among them. As a by-product, we obtain a regularization parameter that can be used to compute such salient segmentations - also on new data sets - even more efficiently.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129970315","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 : 2003-09-17DOI: 10.1109/NNSP.2003.1318014
C. Leroy, J. Bernard, J. Trouilhet
The European Space Agency Planck satellite will be launched in 2007. The goal of this mission is to perform a complete survey of the cosmic microwave background. The high frequency instrument (HFI) on-board Planck would perform all-sky mapping at sub-millimetre and millimetre wavelengths using bolometers cooled at very low temperatures. We have developed a new method able to predict precisely the thermal behaviour of the instrument in order to extract instrumental additive signals due to self-emission by the various cryogenic stages. This article presents a synthesis of the results obtained with neural methods for this modelling problem.
{"title":"Thermal modelling with neural network applied to Planck space mission","authors":"C. Leroy, J. Bernard, J. Trouilhet","doi":"10.1109/NNSP.2003.1318014","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318014","url":null,"abstract":"The European Space Agency Planck satellite will be launched in 2007. The goal of this mission is to perform a complete survey of the cosmic microwave background. The high frequency instrument (HFI) on-board Planck would perform all-sky mapping at sub-millimetre and millimetre wavelengths using bolometers cooled at very low temperatures. We have developed a new method able to predict precisely the thermal behaviour of the instrument in order to extract instrumental additive signals due to self-emission by the various cryogenic stages. This article presents a synthesis of the results obtained with neural methods for this modelling problem.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129175808","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 : 2003-09-17DOI: 10.1109/NNSP.2003.1318026
Y. Wang, J. Xuan, R. Srikanchana, Junying Zhang, Z. Szabo, Z. Bhujwalla, P. Choyke, King C. Li
Functional-molecular imaging techniques promise powerful tools for the visualization and elucidation of important disease-causing physiologic-molecular processes in living tissue. Most applications aim to find temporal-spatial patterns associated with different disease stages. When multiple agents are used, imagery signals often represent a composite of more than one distinct source due to functional-molecular biomarker heterogeneity, independent of spatial resolution. We therefore introduce a hybrid decomposition algorithm, which allows for a computed simultaneous imaging of multiple biomarkers. The method is based on a combination of time-activity curve clustering, pixel subset selection, and independent component analysis. We demonstrate the principle of the approach on an image data set, and we then apply the method to the tumor vascular characterization using dynamic contrast-enhanced magnetic resonance imaging and brain neuro-transporter imaging using dynamic positron emission tomography.
{"title":"Computed simultaneous imaging of multiple biomarkers","authors":"Y. Wang, J. Xuan, R. Srikanchana, Junying Zhang, Z. Szabo, Z. Bhujwalla, P. Choyke, King C. Li","doi":"10.1109/NNSP.2003.1318026","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318026","url":null,"abstract":"Functional-molecular imaging techniques promise powerful tools for the visualization and elucidation of important disease-causing physiologic-molecular processes in living tissue. Most applications aim to find temporal-spatial patterns associated with different disease stages. When multiple agents are used, imagery signals often represent a composite of more than one distinct source due to functional-molecular biomarker heterogeneity, independent of spatial resolution. We therefore introduce a hybrid decomposition algorithm, which allows for a computed simultaneous imaging of multiple biomarkers. The method is based on a combination of time-activity curve clustering, pixel subset selection, and independent component analysis. We demonstrate the principle of the approach on an image data set, and we then apply the method to the tumor vascular characterization using dynamic contrast-enhanced magnetic resonance imaging and brain neuro-transporter imaging using dynamic positron emission tomography.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115344816","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 : 2003-09-17DOI: 10.1109/NNSP.2003.1318003
F. Al-Shahrour, Javier Herrero, Á. Mateos, J. Santoyo, R. Díaz-Uriarte, J. Dopazo
The analysis of genome-scale data from different high throughput techniques usually involves the grouping of genes based on experimental criteria. These groups are a consequence of the biological roles the genes are playing within the cell. Establishing which of these groups are functionally important is essential. Gene ontology terms provide a specialised vocabulary to describe the relevant biological properties of genes. We used a simple procedure to extract terms that are significantly over or under-represented in sets of genes within the context of a genome-scale experiment. Said procedure, which takes the multiple-testing nature of the statistical contrast into account, has been implemented as a Web application, FatiGO, allowing for easy and interactive querying. Several examples demonstrate its application and the type of information that can be extracted. Although a number of genes still lack gene ontology annotations, the results were informative enough to characterise the biological processes in the systems analysed.
{"title":"Using gene ontology on genome-scale studies to find significant associations of biologically relevant terms to groups of genes","authors":"F. Al-Shahrour, Javier Herrero, Á. Mateos, J. Santoyo, R. Díaz-Uriarte, J. Dopazo","doi":"10.1109/NNSP.2003.1318003","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318003","url":null,"abstract":"The analysis of genome-scale data from different high throughput techniques usually involves the grouping of genes based on experimental criteria. These groups are a consequence of the biological roles the genes are playing within the cell. Establishing which of these groups are functionally important is essential. Gene ontology terms provide a specialised vocabulary to describe the relevant biological properties of genes. We used a simple procedure to extract terms that are significantly over or under-represented in sets of genes within the context of a genome-scale experiment. Said procedure, which takes the multiple-testing nature of the statistical contrast into account, has been implemented as a Web application, FatiGO, allowing for easy and interactive querying. Several examples demonstrate its application and the type of information that can be extracted. Although a number of genes still lack gene ontology annotations, the results were informative enough to characterise the biological processes in the systems analysed.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"235 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124581676","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 : 2003-09-17DOI: 10.1109/NNSP.2003.1318031
A. Guerrero-Curieses, R. Alaíz-Rodríguez, Jesús Cid-Sueiro
The design of structures and algorithms for non-MAP multiclass decision problems is discussed in this paper. We propose a parametric family of loss functions that provide the most accurate estimates for the posterior class probabilities near the decision regions. Moreover, we discuss learning algorithms based on the stochastic gradient minimization of these loss functions. We show that these algorithms behave like sample selectors: samples near the decision regions are the most relevant during learning. Experimental results on some real datasets are also provided to show the effectiveness of this approach versus the classical cross entropy (based on a global posterior probability estimation).
{"title":"Loss functions to combine learning and decision in multiclass problems","authors":"A. Guerrero-Curieses, R. Alaíz-Rodríguez, Jesús Cid-Sueiro","doi":"10.1109/NNSP.2003.1318031","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318031","url":null,"abstract":"The design of structures and algorithms for non-MAP multiclass decision problems is discussed in this paper. We propose a parametric family of loss functions that provide the most accurate estimates for the posterior class probabilities near the decision regions. Moreover, we discuss learning algorithms based on the stochastic gradient minimization of these loss functions. We show that these algorithms behave like sample selectors: samples near the decision regions are the most relevant during learning. Experimental results on some real datasets are also provided to show the effectiveness of this approach versus the classical cross entropy (based on a global posterior probability estimation).","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"570 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123041533","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 : 2003-09-17DOI: 10.1109/NNSP.2003.1318041
C. S. Wong, D. Obradovic, N. Madhu
In this paper we address the problem of blind source separation (BSS) in frequency selective multiple-input multiple-output (MIMO) channels, when the only available prior knowledge about the transmitted signals is their mutual statistical independence. The novelty of the paper is two-fold. Firstly, we analytically show that when orthogonal frequency division multiplexing (OFDM) is employed, the original BSS problem is transformed into a set of standard ICA problems with complex mixing matrices. Each ICA problem is associated with one of the orthogonal subcarriers. Secondly, we show that the statistical correlation between the different frequency bins (at each orthogonal subcarrier) can be exploited to avoid the frequency-bin dependent permutation and scaling problems, which are intrinsic to the ICA solution. Our approach is also tested on a realistic channel model.
{"title":"Independent component analysis (ICA) for blind equalization of frequency selective channels","authors":"C. S. Wong, D. Obradovic, N. Madhu","doi":"10.1109/NNSP.2003.1318041","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318041","url":null,"abstract":"In this paper we address the problem of blind source separation (BSS) in frequency selective multiple-input multiple-output (MIMO) channels, when the only available prior knowledge about the transmitted signals is their mutual statistical independence. The novelty of the paper is two-fold. Firstly, we analytically show that when orthogonal frequency division multiplexing (OFDM) is employed, the original BSS problem is transformed into a set of standard ICA problems with complex mixing matrices. Each ICA problem is associated with one of the orthogonal subcarriers. Secondly, we show that the statistical correlation between the different frequency bins (at each orthogonal subcarrier) can be exploited to avoid the frequency-bin dependent permutation and scaling problems, which are intrinsic to the ICA solution. Our approach is also tested on a realistic channel model.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127250086","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 : 2003-09-17DOI: 10.1109/NNSP.2003.1318006
J. A. Berger, S. Hautaniemi, H. Edgren, O. Monni, S. Mitra, O. Yli-Harja, J. Astola
Independent component analysis is a well-known tool for extracting underlying mechanisms from an observed set of parallel data. Identifying such components in breast cancer cell lines, for both copy number and gene expression, is proposed here with the goal of identifying mechanisms that affect the evolution of breast cancer in humans. This paper illustrates how to utilize independent component analysis on cell line data for achieving this goal. After the components were estimated for the well-studied chromosome 17, and then over the entire genome for a set of 14 different breast cancer cell lines, ontological analysis was performed in order to determine common gene functions that are present in each of the independent components.
{"title":"Identifying underlying factors in breast cancer using independent component analysis","authors":"J. A. Berger, S. Hautaniemi, H. Edgren, O. Monni, S. Mitra, O. Yli-Harja, J. Astola","doi":"10.1109/NNSP.2003.1318006","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318006","url":null,"abstract":"Independent component analysis is a well-known tool for extracting underlying mechanisms from an observed set of parallel data. Identifying such components in breast cancer cell lines, for both copy number and gene expression, is proposed here with the goal of identifying mechanisms that affect the evolution of breast cancer in humans. This paper illustrates how to utilize independent component analysis on cell line data for achieving this goal. After the components were estimated for the well-studied chromosome 17, and then over the entire genome for a set of 14 different breast cancer cell lines, ontological analysis was performed in order to determine common gene functions that are present in each of the independent components.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"61 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128733192","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 : 2003-09-17DOI: 10.1109/NNSP.2003.1318024
L. K. Hansen, M. Dyrholm
A linear prediction approach reduces convolutive independent component analysis (ICA) to the following three steps: solution of a set of multivariate linear prediction problems, a linear multivariate deconvolution problem with known matrix coefficients, and finally solution of a conventional instantaneous mixing ICA problem.
{"title":"A prediction matrix approach to convolutive ICA","authors":"L. K. Hansen, M. Dyrholm","doi":"10.1109/NNSP.2003.1318024","DOIUrl":"https://doi.org/10.1109/NNSP.2003.1318024","url":null,"abstract":"A linear prediction approach reduces convolutive independent component analysis (ICA) to the following three steps: solution of a set of multivariate linear prediction problems, a linear multivariate deconvolution problem with known matrix coefficients, and finally solution of a conventional instantaneous mixing ICA problem.","PeriodicalId":315958,"journal":{"name":"2003 IEEE XIII Workshop on Neural Networks for Signal Processing (IEEE Cat. No.03TH8718)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130840585","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：