Charles W. Glover

Reservoir parameter estimation using a hybrid neural network

The accuracy of an artificial neural network (ANN) algorithm is a crucial issue in the estimation of an oil field’s reservoir properties from the log and seismic data. This paper demonstrates the use of the k-fold cross validation technique to obtain confidence bounds on an ANN’s accuracy statistic from a finite sample set. In addition, we also show that an ANN’s classification accuracy is dramatically improved by transforming the ANN’s input feature space to a dimensionally smaller, new input space. The new input space represents a feature space that maximizes the linear separation between classes. Thus, the ANN’s convergence time and accuracy are improved because the ANN must merely find nonlinear perturbations to the starting linear decision boundaries. These techniques for estimating ANN accuracy bounds and feature space transformations are demonstrated on the problem of estimating the sand thickness in an oil field reservoir based only on remotely sensed seismic data. 7 2000 Elsevier Science Ltd. All rights reserved.

Algorithms for recognizing planar polygonal configurations using perspective images

A simplified abstraction of the problem of recognizing planar arrangements of objects using camera pictures taken from unknown positions is considered. A set of polygons in planes is called a planar polygonal configuration. Given perspective images P and Q corresponding to planar polygonal configurations, the matching problem is to determine if P and Q correspond to the same configuration. An optimal theta (n log n) time algorithm is presented to solve this problem, where n is the total number of vertices of polygons in each image. The algorithm is obtained by combining ideas of cross ratios, which are well known to be invariant under perspective projections, and the first fundamental theorem of perspective projections. This algorithm has been implemented and tested with satisfactory results. >

Estimation of reservoir parameter using a hybrid neural network

Estimation of an oil fields reservoir properties using seismic data is a crucial issue. The accuracy of those estimates and the associated uncertainty are also important information. This paper demonstrates the use of the k-fold cross validation technique to obtain confidence bound on an Artificial Neural Networks (ANN) accuracy statistic from a finite sample set. In addition, we also show that an ANNs classification accuracy is dramatically improved by transforming the ANNs input feature space to a dimensionally smaller, new input space. The new input space represents a feature space that maximizes the linear separation between classes. Thus, the ANNs convergence time and accuracy are imporved because the ANN must merely find nonlinear perturbations to the starting linear decision boundaries. These technique for estimating ANN accuracy bounds and feature space transformations are demonstrated on the problem of estimating the sand thickness in an oil field reservoir based only on remotely sensed seismic data.

N-learners problem: fusion of concepts

Given N learners each capable of learning concepts (subsets) in the sense of Valiant (1985), we are interested in combining them using a single fuser. We consider two cases. In open fusion the fuser is given the sample and the hypotheses of the individual learners; we show that a fusion rule can be obtained by formulating this problem as another learning problem. We show sufficiency conditions that ensure the composite system to be better than the best of the individual. Second, in closed fusion the fuser does not have an access to either the training sample or the hypotheses of the individual learners. By using a linear threshold fusion function (of the outputs of individual learners) we show that the composite system can be made better than the best of the statistically independent learners. >

Hybrid neural network and rule-based pattern recognition system capable of self-modification

Abstract not available.

Optimal multicarrier phase-coded waveform design for detection of extended targets

We design a parametric multicarrier phase-coded (MCPC) waveform that achieves the optimal performance in detecting an extended target in the presence of signal-dependent interference. Traditional waveform design techniques provide only the optimal energy spectral density of the transmit waveform and suffer a performance loss in the synthesis process of the time-domain signal. Therefore, we opt for directly designing an MCPC waveform in terms of its time-frequency codes to obtain the optimal detection performance. First, we describe the modeling assumptions considering an extended target buried within the signal-dependent clutter with known power spectral density, and deduce the performance characteristics of the optimal detector. Then, considering an MCPC signal transmission, we express the detection characteristics in terms of phase-codes of the MCPC waveform and propose to optimally design the MCPC signal by maximizing the detection probability. Our numerical results demonstrate that the designed MCPC signal attains the optimal detection performance and requires a lesser computational time than the other parametric waveform design approach.

Frequency adaptability and waveform design for OFDM radar space-time adaptive processing

We propose an adaptive waveform design technique for an orthogonal frequency division multiplexing (OFDM) radar signal employing a space-time adaptive processing (STAP) technique. We observe that there are inherent variabilities of the target and interference responses in the frequency domain. Therefore, the use of an OFDM signal can not only increase the frequency diversity of our system, but also improve the target detectability by adaptively modifying the OFDM-coefficients in order to exploit the frequency-variabilities of the scenario. First, we formulate a realistic OFDM-STAP measurement model considering the sparse nature of the target and interference spectra in the spatio-temporal domain. Then, we show that the optimal STAP-filter weight-vector is equal to the generalized eigenvector corresponding to the minimum generalized eigenvalue of the interference and target covariance matrices. With numerical examples we demonstrate that the resultant OFDM-STAP filter-weights are adaptable to the frequency-variabilities of the target and interference responses, in addition to the spatio-temporal variabilities. Hence, by better utilizing the frequency variabilities, we propose an adaptive OFDM-waveform design technique, and consequently gain a significant amount of STAP-performance improvement.

Does extra information always help in data fusion

Adding new sensor metric information into a data fusion process does not always improve performance and can sometimes produce poorer results. References 1 and 2 used examples to show that - in some instances and contrary to expectation - adding new information resulted in poorer rather than improved performance, even though the information itself was correct. Correct is being used here to describe data that may be in error because of sensor deficiencies but whose error characteristics are accurately described and known to the fusion process. In other words, the fusion process is not being lied to be misrepresentation of the data quality. In this sense, an individual data point may be inaccurate, but the fusion process is capable of properly weighting that point in an optimal sensors that its statistical inaccuracy does not damage the final product any more than a data point from a better sensor that has less statistical inaccuracy. In a multiple-sensor fusion process, these kinds of result have been cited as reasons for not using data from poorer quality sensors for fear of diluting the performance of the better quality sensors. This paper explores the counterintuitive findings for these referenced examples and evaluates under what conditions lesser quality sensor or sensor that mistakenly overestimate their own data quality should be allowed to contribute to a sensor fusion process.

Accuracy estimation for supervised learning algorithms

This paper illustrates and discusses the relative merits of three methods--k-fold Cross Validation, Error Bounds, and Incremental Halting Test--to estimate the accuracy of a supervised learning algorithm. For each of the three methods we point out the problem they address, some of the important assumptions that they are based on, and illustrate them through an example. Finally, we discuss the relative advantages and disadvantages of each method.

N-learners Problem: Fusion Of Concepts

We are given N learners each capable of learning concepts (subsets) of a domain set x in the sense of Vdiant, i.e. for any c E C c 2x, given a finite set of examples of the form < 2 1 , A d c ( x l ) >; < 5 2 , Mc(zz ) >;. . . ; < 51, llfc(x,) > generated according to an unknown probability distribution Px on X, each learner produces a close approximation to c with a high probability. We are interested in combining the N learners using a single f u s e r or consolidator, We consider the paradigm of passive fusion, where each learner is first trained with the sample without the influence of the consolidator. The composi te s y s t e m is constituted by the fuser and the individual learners. We consider two cases: open and closed fusion. In o p e n f u s i o n the fuser is given the sample and the hypotheses of the individual learners; we show that the fusion rule can be obtained by formulating this problem as another learning problem. For the case all individual learners are trained with the same sample, we show sufficiency conditions that ensure the composite system to be better than the best of the individual: the hypothesis space of the consolidator (a) satisfies the i so la t ion property of degree at least N , and (b) has Vapnik-Chervonenkis diincnsion less than or equal to that of every individual learner. If individual lea,rners are trained by independently generated samples, we obtain a rnuch weaker bound on the VC-dimension of the hypothesis space of the fuser. Second, in closed fusion the fuser does not have an access to either the training sample or the hypotheses of thc individual learners. By suitably designing a linear threshold function of the outputs of individual learners, we show that the composite system can be made bctter than the best of the learners.