eLearning

Maximum mutual information speaker adapted training with semi-tied covariance matrices

J. McDonough; A. Waibel Acoustics, Speech, and Signal Processing, 2003. Proceedings. (ICASSP '03). 2003 IEEE International Conference on, 2003

We present re-estimation formulae for semi-tied covariance (STC) transformation matrices based on a maximum mutual information (MMI) criterion. These re-estimation formulae are different from those that have appeared previously in the literature. Moreover, we present a positive definiteness criterion with which the regularization constant present in all NMI re- estimation formulae can be reliably set to provide both consistent improvements ...

Two--dimensional vector modelling of image random field using artificial neural networks

1993 IEEE International Symposium on Circuits and Systems, 1993

First Page of the Article 

An efficient method for simulating fractional stable motion

Wei Biao Wu; G. Michailidis; Danlu Zhang Proceedings of the Winter Simulation Conference, 2002

An efficient methodology for simulating paths of fractional stable motion is presented. The proposed approach is based on invariance principles for linear processes. A detailed analysis of the error terms involved is given and the performance of the method is assessed through an extensive simulation study.

On Parameter Estimation Using Nonparametric Noise Models

K. Mahata; R. Pintelon; J. Schoukens IEEE Transactions on Automatic Control, 2006

Fitting multidimensional parametric models in frequency domain using nonparametric noise models is considered in this paper. A nonparametric estimate of the noise statistics is obtained from a finite number of independent data sets. The estimated noise model is then substituted for the the true noise covariance matrix in the maximum likelihood loss function to obtain suboptimal parameter estimates. The goal ...

Joint source-channel decoding of speech spectrum parameters over erasure channels using Gaussian mixture models

A. D. Subramaniam; W. R. Gardner; B. D. Rao Acoustics, Speech, and Signal Processing, 2003. Proceedings. (ICASSP '03). 2003 IEEE International Conference on, 2003

A joint source-channel decoding scheme that improves the performance of conventional channel decoders over erasure channels by exploiting the cross- correlation between successive speech frames is presented. Speech spectrum parameters are quantized using the scheme presented in Subramaniam and Rao (2001). The joint probability density function (PDF) of the spectrum parameters of successive speech frames is modelled using a Gaussian ...

More eLearning Resources

IEEE-USA E-Books

• Vector Radiative Transfer

  Chapter 2 introduces the vector radiative transfer (VRT) theory of random media. It provides the scattering, absorption and extinction coefficients and the phase matrix of non-spherical scatterers in natural media. The first-order Mueller matrix solution of VRT for vegetation canopy model is derived. Polarization indexes, eigen-analysis and entropy are presented. Statistics of multi-look SAR images and covariance matrix are also investigated.

• Some Probability and Stochastic Convergence Fundamentals

  This chapter contains sections titled: Notations and Definitions The Covariance Matrix of a Function of a Random Variable Sample Variables Mixing Random Variables Preliminary Example Definitions of Stochastic Limits Interrelations between Stochastic Limits Properties of Stochastic Limits Laws of Large Numbers Central Limit Theorems Properties of Estimators Cramér-Rao Lower Bound How to Prove Asymptotic Properties of Estimators? Pitfalls Preliminary Example - Continued Properties of the Noise after a Discrete Fourier Transform Exercises Appendixes

• Evaluating Multisensor Classification Performance with Bayesian Networks

  This chapter presents a new analytical approach for quantifying the long¿¿?run performance of a multisensor, discrete¿¿?state classification system, under the assumption of independent, asynchronous measurements, and consolidates earlier research. It addresses the problem of evaluating the classification performance of both single¿¿? and multisensor classification systems, where one or more sensors observe repeated measurements of a target's features/attributes and compute posterior probability estimates to aid in target identification. The chapter develops a new analytical approach for off¿¿?line evaluation of the long¿¿?run classification performance of a single¿¿? or a multisensor system for the case of independent sensor measurement. It defines a new approach to quantify the classification performance based on the global classification matrix (GCM) of a classification system. The GCM is analogous to the covariance matrix used in kinematic performance evaluation and is an off¿¿?line measure that does not depend on the actual sensor measurements.

• A Serial Approach to Handling High-Dimensional Measurements in the Sigma-Point Kalman Filter

  Pose estimation is a critical skill in mobile robotics and is often accomplished using onboard sensors and a Kalman filter estimation technique. For systems to run online, computational efficiency of the filter design is crucial, especially when faced with limited computing resources. In this paper, we present a novel approach to serially process high-dimensional measurements in the Sigma-Point Kalman Filter (SPKF), in order to achieve a low computational cost that is linear is the measurement dimension. Although the concept of serially processing measurements has been around for quite some time in the context of the Extended Kalman Filter (EKF), few have considered this approach with the SPKF. At first glance, it may be tempting to apply the SPKF update step serially. However, we prove that without re-drawing sigma points, this 'naive' approach cannot guarantee the positive-definiteness of the state covariance matrix (not the case for the EKF). We then introduce a novel method for the Sigma-Point Kalman Filter to process high-dimensional, uncorrelated measurements serially that is algebraically equivalent to processing the measurements in parallel, but still achieves a computational cost linear in the measurement dimension.

• Analysis of Contour Motions

  A reliable motion estimation algorithm must function under a wide range of conditions. One regime, which we consider here, is the case of moving objects with contours but no visible texture. Tracking distinctive features such as corners can disambiguate the motion of contours, but spurious features such as T-junctions can be badly misleading. It is difficult to determine the reliability of motion from local measurements, since a full rank covariance matrix can result from both real and spurious features. We propose a novel approach that avoids these points altogether, and derives global motion estimates by utilizing information from three levels of contour analysis: edgelets, boundary fragments and contours. Boundary fragment are chains of orientated edgelets, for which we derive motion estimates from local evidence. The uncertainties of the local estimates are disambiguated after the boundary fragments are properly grouped into contours. The grouping is done by constructing a graphical model and marginalizing it using importance sampling. We propose two equivalent representations in this graphical model, reversible switch variables attached to the ends of fragments and fragment chains, to capture both local and global statistics of boundaries. Our system is successfully applied to both synthetic and real video sequences containing high-contrast boundaries and textureless regions. The system produces good motion estimates along with properly grouped and completed contours.

• Approaches to High-Dimensional Covariance and Precision Matrix Estimations

  This chapter introduces several recent developments for estimating large covariance and precision matrices without assuming the covariance matrix to be sparse. It explains two methods for covariance estimation: namely covariance estimation via factor analysis, and precision Matrix Estimation and Graphical Models. The low rank plus sparse representation holds on the population covariance matrix. The chapter presents several applications of these methods, including graph estimation for gene expression data, and several financial applications. It then shows how estimating covariance matrices of high- dimensional asset excess returns play a central role in applications of portfolio allocations and in risk management. The chapter explains the factor pricing model, which is one of the most fundamental results in finance. It elucidates estimating risks of large portfolios and large panel test of factor pricing models. The chapter illustrates the recent developments of efficient estimations in panel data models.

• Optimal Polarizations for Radar

  This chapter contains sections titled: Antenna Selection Criteria Lagrange Multipliers Maximum Power Power Contrast: Backscattering Iterative Procedure for Maximizing Power Contrast The Backscattering Covariance Matrix The Bistatic Covariance Matrix Maximizing Power Contrast by Matrix Decomposition Optimization with the Graves Matrix References Problems

• MeanSquared Error and Threshold SNR Prediction of MaximumLikelihood Signal Parameter Estimation With Estimated Colored Noise Covariances

  An interval error-based method (MIE) of predicting mean squared error (MSE) performance of maximum-likelihood estimators (MLEs) is extended to the case of signal parameter estimation requiring intermediate estimation of an unknown colored noise covariance matrix; an intermediate step central to adaptive array detection and parameter estimation. The successful application of MIE requires good approximations of two quantities: 1) interval error probabilities and 2) asymptotic (SNR local MSE performance of the MLE. Exact general expressions for the pairwise error probabilities that include the effects of signal model mismatch are derived herein, that in conjunction with the Union Bound provide accurate prediction of the required interval error probabilities. The Cramér-Ran Bound (CRB) often provides adequate prediction of the asymptotic local MSE performance of MLE. The signal parameters, however, are decoupled from the colored noise parameters in the Fisher Information Matrix for the deterministic signal model, rendering the CRB incapable of reflecting loss due to colored noise covariance estimation. A new modification of the CRB involving a complex central beta random variable different from, but analogous to the Reed, Mallett, and Brennan beta loss factor provides a working solution to this problem, facilitating MSE prediction well into the threshold region with remarkable accuracy.

• Utilization of Modified Polar Coordinates for BearingsOnly Tracking

  Previous studies have shown that the Cartesian coordinate extended Kalman filter exhibits unstable behavior characteristics when utilized for bearings- only target motion analysis (TMA). In contrast. fonnulating the TMA estimation problem in modified polar (MP) coordinates leads to an extended Kalman filter which is both stable and asymptotically unbiased. Exact state equations for the MP filter are derived without imposing any restrictions on own-ship motion; thus, prediction accuracy inherent in the traditional Cartesian formulation is completely preserved. In addition, these equations reveal that MP coordinates are well-suited for bearings-only TMA because they automatically decouple observable and unobservable components of the estimated state vector.Such decoupling is shown to prevent covariance matrix ill- condttioning, which is the primary cause of filter instability. Further investigation also confirms that the MP state estimates are asymptotically unbiased. Realistic simulation data are presented to support these findings and to compare algorithm performance with respect to the Cramer-Rao lower bound (ideal) as well as the Cartesian and pseudolinear filters.

• Differential Entropic Clustering of Multivariate Gaussians

  Gaussian data is pervasive and many learning algorithms (e.g., k-means) model their inputs as a single sample drawn from a multivariate Gaussian. However, in many real-life settings, each input object is best described by multiple samples drawn from a multivariate Gaussian. Such data can arise, for example, in a movie review database where each movie is rated by several users, or in time-series domains such as sensor networks. Here, each input can be naturally described by both a mean vector and covariance matrix which parameterize the Gaussian distribution. In this paper, we consider the problem of clustering such input objects, each represented as a multivariate Gaussian. We formulate the problem using an information theoretic approach and draw several interesting theoretical connections to Bregman divergences and also Bregman matrix divergences. We evaluate our method across several domains, including synthetic data, sensor network data, and a statistical debugging application.