The effect of the near surface on seismic land data can be so severe that static corrections are ... more The effect of the near surface on seismic land data can be so severe that static corrections are insufficient. Full-waveform inversion followed by redatuming may be an alternative, but inversion will work only if the starting model is sufficiently close to the true model. As a first step toward determining a viscoelastic near-surface model, we assume that existing methods can provide a horizontally layered velocity and density model. Because near-surface attenuation is strongest, we propose a method to estimate the P-wave attenuation based on viscoacoustic finite-difference modeling. We compare energy decay along traveltime curves of reflection and refraction events in the modeled and observed seismic data for a range of attenuation parameters. The best match provides an estimate of the attenuation. First, we estimate only the attenuation of the top layer and study the sensitivity to depth and velocity perturbations. Then, we consider multiple layers. We apply the method to synthetic and real data and investigate the effect of source wavelet and topography. The method is robust against depth and velocity perturbations smaller than 10%. The results are sensitive to the source wavelet. Incorporating the surface topography in the computed traveltimes reduces the uncertainty of the attenuation estimates, especially for deeper layers.
Frequency-to-time transformations are of interest to controlled-source electromagnetic methods wh... more Frequency-to-time transformations are of interest to controlled-source electromagnetic methods when time-domain data are inverted for a subsurface resistivity model by numerical frequency-domain modeling at a selected, small number of frequencies whereas the data misfit is determined in the time domain. We propose an efficient, Prony-type method using frequency-domain diffusive-field basis functions for which the time-domain equivalents are known. Diffusive fields are characterized by an exponential part whose argument is proportional to the square root of frequency and a part that is polynomial in integer powers of the square root of frequency. Data at a limited number of frequencies suffice for the transformation back to the time. In the exponential part, several diffusion-time values must be chosen. Once a suitable range of diffusion-time values are found, the method is quite robust in the number of values used. The highest power in the polynomial part can be determined from the source and receiver type. When the frequency-domain data are accurately approximated by the basis functions, the timedomain result is also accurate. This method is accurate over a wider time range than other methods and has the correct late-time asymptotic behavior. The method works well for data computed for layered and 3D subsurface models.
In marine time-domain controlled-source electromagnetics (CSEM), there are two different acquisit... more In marine time-domain controlled-source electromagnetics (CSEM), there are two different acquisition methods: with horizontal sources for fast and simple data acquisition or with vertical sources for minimizing the effects of the airwave. Illustrations of the electric field as a function of space and time for various source antenna orientations, based on analytical formulation of the electric field in two half-spaces, provide insights into the properties of the airwave and the nature of diffuse electric fields. Observing the development of the electric field over time and space reveals that diffusive fields exhibit directionality. Therefore, techniques that have thus far mostly been applied to wavefields can be adapted for CSEM. Examples range from the well-known up-down decomposition to beam steering. Vertical sources have the advantage of not creating an airwave. On the other hand, it is quite difficult to achieve perfect verticality of the source antenna. Results, using a numerically modeled data set to analyze the impact of the airwave on a signal from a subsurface reservoir in the case of a slightly dipping vertical source, indicate that already for a dip of 0:05 , the airwave contributes 20% to the complete electric field in our configuration of reservoir depth, water thickness, and conductivity values.
International Journal of Computer Mathematics, 2011
In this paper, we generalize the hard clustering paradigm. While in this paradigm a data set is s... more In this paper, we generalize the hard clustering paradigm. While in this paradigm a data set is subdivided into disjoint clusters, we allow different clusters to have a nonempty intersection. The concept of hard clustering is then analysed in this general setting, and we show which specific properties hard clusterings possess in comparison to more general clusterings. We also introduce the concept of equivalent clusterings and show that in the case of hard clusterings equivalence and equality coincide. However, if more general clusterings are considered, these two concepts differ, and this implies the undesired fact that equivalent clusterings can have different representations in the traditional view on clustering. We show how a matrix representation can solve this representation problem.
International Conference on Computer and Automation Engineering, 2010
We propose a measure for the validation of clusterings of gene expression data. This measure also... more We propose a measure for the validation of clusterings of gene expression data. This measure also useful to estimate missing gene expression levels, based the similarity information contained in a given clustering. It is shown that this measure is an improvement over the figure of merit, an existing validation measure especially developed for clusterings of gene expression data.
Clustering is an important approach in the analysis of biological data, and often a first step to... more Clustering is an important approach in the analysis of biological data, and often a first step to identify interesting patterns of coexpression in gene expression data. Because of the high complexity and diversity of gene expression data, many genes cannot be easily assigned to a cluster, but even if the dissimilarity of these genes with all other gene groups is large, they will finally be forced to become member of a cluster. In this paper we show how to detect such elements, called unstable elements. We have developed an approach for iterative clustering algorithms in which unstable elements are deleted, making the iterative algorithm less dependent on initial centers. Although the approach is unsupervised, it is less likely that the clusters into which the reduced data set is subdivided contain false positives. This clustering yields a more differentiated approach for biological data, since the cluster analysis is divided into two parts: the pruned data set is divided into highly consistent clusters in an unsupervised way and the removed, unstable elements for which no meaningful cluster exists in unsupervised terms can be given a cluster with the use of biological knowledge and information about the likelihood of cluster membership. We illustrate our framework on both an artificial and real biological data set. a Corresponding author: [email protected]
A survey on the application of recurrent neural networks to statistical language modeling
Computer Speech & Language, 2015
ABSTRACT In this paper we present a survey on the application of recurrent neural networks to the... more ABSTRACT In this paper we present a survey on the application of recurrent neural networks to the task of statistical language modeling. Although it has been shown that these models obtain good performance on this task, often superior to other state-of-the-art techniques, they suffer from some important drawbacks, including a very long training time and limitations on the number of context words that can be taken into account in practice. Recent extensions to recurrent neural network models have been developed in an attempt to address these drawbacks. This paper gives an overview of the most important extensions. Each technique is described and its performance on statistical language modeling, as described in the existing literature, is discussed. Our structured overview makes it possible to detect the most promising techniques in the field of recurrent neural networks, applied to language modeling, but it also highlights the techniques for which further research is required.
2010 The 2nd International Conference on Computer and Automation Engineering (ICCAE), 2010
We propose a measure for the validation of clusterings of gene expression data. This measure also... more We propose a measure for the validation of clusterings of gene expression data. This measure also useful to estimate missing gene expression levels, based the similarity information contained in a given clustering. It is shown that this measure is an improvement over the figure of merit, an existing validation measure especially developed for clusterings of gene expression data.
Initialization Dependence of Clustering Algorithms
It is well known that the clusters produced by a clustering algorithm depend on the chosen initia... more It is well known that the clusters produced by a clustering algorithm depend on the chosen initial centers. In this paper we present a measure for the degree to which a given clustering algorithm depends on the choice of initial centers, for a given data set. This measure is calculated for four well-known offline clustering algorithms (k-means Forgy, k-means Hartigan, k-means Lloyd and fuzzy c-means), for five benchmark data sets. The measure is also calculated for ECM, an online algorithm that does not require the number of initial centers as input, but for which the resulting clusters can depend on the order that the input arrives. Our main finding is that this initialization dependence measure can also be used to determine the optimal number of clusters.
In this paper we give a general definition for the concept 'optimal clustering' which is applicab... more In this paper we give a general definition for the concept 'optimal clustering' which is applicable to overlapping clusterings. Overlapping clusterings are a generalization of hard clusterings and their structure is formally developed in this paper. It is generally assumed that the domain of clustering is too heuristic to develop a general, i.e. axiomatic, definition for an optimal clustering. It is shown, however, that such a definition can be given within the domain of overlapping clusterings, using the new concept of dual clustering developed in this paper. A second concept that underlies our definition of optimal clustering is the average clustering, also playing an important role in the domain of cluster ensembles. Using the general concepts discussed in this paper, it is then shown that under some conditions it is assured that the final hard clustering extracted by majority vote from a given set of clusterings, is optimal over all hard clusterings.
Initialization dependence of clustering algorithms
Advances in Neuro- …, 2009
It is well known that the clusters produced by a clustering algorithm depend on the chosen initia... more It is well known that the clusters produced by a clustering algorithm depend on the chosen initial centers. In this paper we present a measure for the degree to which a given clustering algorithm depends on the choice of initial centers, for a given data set. This measure is ...
The effect of the near surface on seismic land data can be so severe that static corrections are ... more The effect of the near surface on seismic land data can be so severe that static corrections are insufficient. Full-waveform inversion followed by redatuming may be an alternative, but inversion will work only if the starting model is sufficiently close to the true model. As a first step toward determining a viscoelastic near-surface model, we assume that existing methods can provide a horizontally layered velocity and density model. Because near-surface attenuation is strongest, we propose a method to estimate the P-wave attenuation based on viscoacoustic finite-difference modeling. We compare energy decay along traveltime curves of reflection and refraction events in the modeled and observed seismic data for a range of attenuation parameters. The best match provides an estimate of the attenuation. First, we estimate only the attenuation of the top layer and study the sensitivity to depth and velocity perturbations. Then, we consider multiple layers. We apply the method to synthetic and real data and investigate the effect of source wavelet and topography. The method is robust against depth and velocity perturbations smaller than 10%. The results are sensitive to the source wavelet. Incorporating the surface topography in the computed traveltimes reduces the uncertainty of the attenuation estimates, especially for deeper layers.
Frequency-to-time transformations are of interest to controlled-source electromagnetic methods wh... more Frequency-to-time transformations are of interest to controlled-source electromagnetic methods when time-domain data are inverted for a subsurface resistivity model by numerical frequency-domain modeling at a selected, small number of frequencies whereas the data misfit is determined in the time domain. We propose an efficient, Prony-type method using frequency-domain diffusive-field basis functions for which the time-domain equivalents are known. Diffusive fields are characterized by an exponential part whose argument is proportional to the square root of frequency and a part that is polynomial in integer powers of the square root of frequency. Data at a limited number of frequencies suffice for the transformation back to the time. In the exponential part, several diffusion-time values must be chosen. Once a suitable range of diffusion-time values are found, the method is quite robust in the number of values used. The highest power in the polynomial part can be determined from the source and receiver type. When the frequency-domain data are accurately approximated by the basis functions, the timedomain result is also accurate. This method is accurate over a wider time range than other methods and has the correct late-time asymptotic behavior. The method works well for data computed for layered and 3D subsurface models.
In marine time-domain controlled-source electromagnetics (CSEM), there are two different acquisit... more In marine time-domain controlled-source electromagnetics (CSEM), there are two different acquisition methods: with horizontal sources for fast and simple data acquisition or with vertical sources for minimizing the effects of the airwave. Illustrations of the electric field as a function of space and time for various source antenna orientations, based on analytical formulation of the electric field in two half-spaces, provide insights into the properties of the airwave and the nature of diffuse electric fields. Observing the development of the electric field over time and space reveals that diffusive fields exhibit directionality. Therefore, techniques that have thus far mostly been applied to wavefields can be adapted for CSEM. Examples range from the well-known up-down decomposition to beam steering. Vertical sources have the advantage of not creating an airwave. On the other hand, it is quite difficult to achieve perfect verticality of the source antenna. Results, using a numerically modeled data set to analyze the impact of the airwave on a signal from a subsurface reservoir in the case of a slightly dipping vertical source, indicate that already for a dip of 0:05 , the airwave contributes 20% to the complete electric field in our configuration of reservoir depth, water thickness, and conductivity values.
International Journal of Computer Mathematics, 2011
In this paper, we generalize the hard clustering paradigm. While in this paradigm a data set is s... more In this paper, we generalize the hard clustering paradigm. While in this paradigm a data set is subdivided into disjoint clusters, we allow different clusters to have a nonempty intersection. The concept of hard clustering is then analysed in this general setting, and we show which specific properties hard clusterings possess in comparison to more general clusterings. We also introduce the concept of equivalent clusterings and show that in the case of hard clusterings equivalence and equality coincide. However, if more general clusterings are considered, these two concepts differ, and this implies the undesired fact that equivalent clusterings can have different representations in the traditional view on clustering. We show how a matrix representation can solve this representation problem.
International Conference on Computer and Automation Engineering, 2010
We propose a measure for the validation of clusterings of gene expression data. This measure also... more We propose a measure for the validation of clusterings of gene expression data. This measure also useful to estimate missing gene expression levels, based the similarity information contained in a given clustering. It is shown that this measure is an improvement over the figure of merit, an existing validation measure especially developed for clusterings of gene expression data.
Clustering is an important approach in the analysis of biological data, and often a first step to... more Clustering is an important approach in the analysis of biological data, and often a first step to identify interesting patterns of coexpression in gene expression data. Because of the high complexity and diversity of gene expression data, many genes cannot be easily assigned to a cluster, but even if the dissimilarity of these genes with all other gene groups is large, they will finally be forced to become member of a cluster. In this paper we show how to detect such elements, called unstable elements. We have developed an approach for iterative clustering algorithms in which unstable elements are deleted, making the iterative algorithm less dependent on initial centers. Although the approach is unsupervised, it is less likely that the clusters into which the reduced data set is subdivided contain false positives. This clustering yields a more differentiated approach for biological data, since the cluster analysis is divided into two parts: the pruned data set is divided into highly consistent clusters in an unsupervised way and the removed, unstable elements for which no meaningful cluster exists in unsupervised terms can be given a cluster with the use of biological knowledge and information about the likelihood of cluster membership. We illustrate our framework on both an artificial and real biological data set. a Corresponding author: [email protected]
A survey on the application of recurrent neural networks to statistical language modeling
Computer Speech & Language, 2015
ABSTRACT In this paper we present a survey on the application of recurrent neural networks to the... more ABSTRACT In this paper we present a survey on the application of recurrent neural networks to the task of statistical language modeling. Although it has been shown that these models obtain good performance on this task, often superior to other state-of-the-art techniques, they suffer from some important drawbacks, including a very long training time and limitations on the number of context words that can be taken into account in practice. Recent extensions to recurrent neural network models have been developed in an attempt to address these drawbacks. This paper gives an overview of the most important extensions. Each technique is described and its performance on statistical language modeling, as described in the existing literature, is discussed. Our structured overview makes it possible to detect the most promising techniques in the field of recurrent neural networks, applied to language modeling, but it also highlights the techniques for which further research is required.
2010 The 2nd International Conference on Computer and Automation Engineering (ICCAE), 2010
We propose a measure for the validation of clusterings of gene expression data. This measure also... more We propose a measure for the validation of clusterings of gene expression data. This measure also useful to estimate missing gene expression levels, based the similarity information contained in a given clustering. It is shown that this measure is an improvement over the figure of merit, an existing validation measure especially developed for clusterings of gene expression data.
Initialization Dependence of Clustering Algorithms
It is well known that the clusters produced by a clustering algorithm depend on the chosen initia... more It is well known that the clusters produced by a clustering algorithm depend on the chosen initial centers. In this paper we present a measure for the degree to which a given clustering algorithm depends on the choice of initial centers, for a given data set. This measure is calculated for four well-known offline clustering algorithms (k-means Forgy, k-means Hartigan, k-means Lloyd and fuzzy c-means), for five benchmark data sets. The measure is also calculated for ECM, an online algorithm that does not require the number of initial centers as input, but for which the resulting clusters can depend on the order that the input arrives. Our main finding is that this initialization dependence measure can also be used to determine the optimal number of clusters.
In this paper we give a general definition for the concept 'optimal clustering' which is applicab... more In this paper we give a general definition for the concept 'optimal clustering' which is applicable to overlapping clusterings. Overlapping clusterings are a generalization of hard clusterings and their structure is formally developed in this paper. It is generally assumed that the domain of clustering is too heuristic to develop a general, i.e. axiomatic, definition for an optimal clustering. It is shown, however, that such a definition can be given within the domain of overlapping clusterings, using the new concept of dual clustering developed in this paper. A second concept that underlies our definition of optimal clustering is the average clustering, also playing an important role in the domain of cluster ensembles. Using the general concepts discussed in this paper, it is then shown that under some conditions it is assured that the final hard clustering extracted by majority vote from a given set of clusterings, is optimal over all hard clusterings.
Initialization dependence of clustering algorithms
Advances in Neuro- …, 2009
It is well known that the clusters produced by a clustering algorithm depend on the chosen initia... more It is well known that the clusters produced by a clustering algorithm depend on the chosen initial centers. In this paper we present a measure for the degree to which a given clustering algorithm depends on the choice of initial centers, for a given data set. This measure is ...
Uploads
Papers by Wim De Mulder