Squared Extrapolation Methods (SQUAREM): A New Class of Simple and Efficient Numerical Schemes for Accelerating the Convergence of the EM Algorithm

We derive a new class of iterative schemes for accelerating the convergence of the EM algorithm, by exploiting the connection between fixed point iterations and extrapolation methods. First, we present a general formulation of one-step iterative schemes, which are obtained by cycling with the extrapolation methods. We, then square the one-step schemes to obtain the new class of methods, which we call SQUAREM. Squaring a one-step iterative scheme is simply applying it twice within each cycle of the extrapolation method. Here we focus on the first order or rank-one extrapolation methods for two reasons, (1) simplicity, and (2) computational efficiency. In particular, we study two first order extrapolation methods, the reduced rank extrapolation (RRE1) and minimal polynomial extrapolation (MPE1). The convergence of the new schemes, both one-step and squared, is non-monotonic with respect to the residual norm. The first order one-step and SQUAREM schemes are linearly convergent, like the EM algorithm but they have a faster rate of convergence. We demonstrate, through five different examples, the effectiveness of the first order SQUAREM schemes, SqRRE1 and SqMPE1, in accelerating the EM algorithm. The SQUAREM schemes are also shown to be vastly superior to their one-step counterparts, RRE1 and MPE1, in terms of computational efficiency. The proposed extrapolation schemes can fail due to the numerical problems of stagnation and near breakdown. We have developed a new hybrid iterative scheme that combines the RRE1 and MPE1 schemes in such a manner that it overcomes both stagnation and near breakdown. The squared first order hybrid scheme, SqHyb1, emerges as the iterative scheme of choice based on our numerical experiments. It combines the fast convergence of the SqMPE1, while avoiding near breakdowns, with the stability of SqRRE1, while avoiding stagnations. The SQUAREM methods can be incorporated very easily into an existing EM algorithm. They only require the basic EM step for their implementation and do not require any other auxiliary quantities such as the complete data log likelihood, and its gradient or hessian. They are an attractive option in problems with a very large number of parameters, and in problems where the statistical model is complex, the EM algorithm is slow and each EM step is computationally demanding.

Numerical solution of stiff ordinary differential equations using collocation methods /

Bibliography: p. 79-80.

Methods for automatic error analysis of numerical algorithms /

Vita.

Numerical studies of the Stone algorithm and comparisons with Alternating Direction Implicit methods /

Thesis (M.S.)--University of Illinois, 1970.

The approximate numerical solution of boundary value problems by relaxation methods,

"Contract N7 onr-358, T. O. I., NR-041-032."

Numerical modeling of 11.1 T MRI of a human head using MoM/FDTD methods

In a recent study, severe distortions in the proton images of an excised, fixed, human brain in an 11.1 Tesla/40 cm MR instrument have been observed, and the effect modeled on phantom images using a finite difference time domain (FDTD) model. in the present study, we extend these simulations to that of a complete human head, employing a hybrid FDTD and method of moments (MoM) approach, which provides a validated method for simulating biological samples in coil structures. The effect of fixative on the image distortions is explored. importantly, temperature distributions within the head are also simulated using a bioheat method based on parameters derived from the electromagnetic simulations. The MoM/FDTD simulations confirm that the transverse magnetic field (B,) from a ReCav resonator exhibits good homogeneity in air but strong inhomogeneity when loaded with the head with or without fixative. The fixative serves to increase the distortions, but they are still significant for the in vivo simulations. The simulated signal intensity (SI) distribution within the sample confirm the distortions in the experimental images are caused by the complex interactions of the incident electromagnetic fields with tissue, which is heterogeneous in terms of conductivity and permittivity. The temperature distribution is likewise heterogeneous, raising concerns regarding hot spot generation in the sample that may exceed acceptable levels in future in vivo studies. As human imaging at 11.1 T is some time away, simulations are important in terms of predicting potential safety issues as well as evaluating practical concerns about the quality of images. Simulation on a whole human head at 11.1 T implies the wave behavior presents significant engineering challenges for ultra-high-field (UHF) MRI. Novel strategies will have to be employed in imaging technique and resonator design for UHF MRI to achieve the theoretical signal-to-noise ratio (SNR) improvements it offers over lower field systems. (C) 2005 Wiley Periodicals, Inc.

Comparison of BER estimation methods in numerical simulation of 40 Gbit/s RZ-DPSK transmission

This study analyzes the validity of different Q-factor models in the BER estimation in RZ-DPSK transmission at 40 Gb/s channel rate. The impact of the duty cycle of the carrier pulses on the accuracy of the BER estimates through the different models has also been studied.

Comparative analysis of BER estimation methods in numerical simulation of 40-Gb/s RZ-DPSK transmission with in-line SOAs

Applying direct error counting, we compare the accuracy and evaluate the validity of different available numerical approaches to the estimation of the bit-error rate (BER) in 40-Gb/s return-to-zero differential phase-shift-keying transmission. As a particular example, we consider a system with in-line semiconductor optical amplifiers. We demonstrate that none of the existing models has an absolute superiority over the others. We also reveal the impact of the duty cycle on the accuracy of the BER estimates through the differently introduced Q-factors.

A survey of methods of numerical approximation to functions

DUE TO COPYRIGHT RESTRICTIONS ONLY AVAILABLE FOR CONSULTATION AT ASTON UNIVERSITY LIBRARY AND INFORMATION SERVICES WITH PRIOR ARRANGEMENT

Comparison of BER estimation methods in numerical simulation of 40 Gbit/s RZ-DPSK transmission

This study analyzes the validity of different Q-factor models in the BER estimation in RZ-DPSK transmission at 40 Gb/s channel rate. The impact of the duty cycle of the carrier pulses on the accuracy of the BER estimates through the different models has also been studied.

Comparative analysis of BER estimation methods in numerical simulation of 40-Gb/s RZ-DPSK transmission with in-line SOAs

Applying direct error counting, we compare the accuracy and evaluate the validity of different available numerical approaches to the estimation of the bit-error rate (BER) in 40-Gb/s return-to-zero differential phase-shift-keying transmission. As a particular example, we consider a system with in-line semiconductor optical amplifiers. We demonstrate that none of the existing models has an absolute superiority over the others. We also reveal the impact of the duty cycle on the accuracy of the BER estimates through the differently introduced Q-factors. © 2007 IEEE.

Comparison between Numerical and Simulation Methods for Age-dependent Branching Models with Immigration

2000 Mathematics Subject Classification: primary: 60J80, 60J85, secondary: 62M09, 92D40

Comparison of numerical bit error rate estimation methods in 112Gbs QPSK CO-OFDM transmission

We demonstrate an accurate BER estimation method for QPSK CO-OFDM transmission based on the probability density function of the received QPSK symbols. Using a 112Gbs QPSK CO-OFDM transmission as an example, we show that this method offers the most accurate estimate of the system's performance in comparison with other known approaches.

Pseudospectral methods and numerical continuation for the analysis os structured population models.

191 p.

Esophageal Ulcer in Brazilian Patients with HIV: Prevalence and Comparative Analysis Among Diagnostic Methods

Esophageal ulcer (EU) represents an important comorbidity in AIDS. We evaluated the prevalence of EU, the accuracy of the endoscopic and histologic methods used to investigate viral EU in HIV-positive Brazilian patients and the numerical relevance of tissue sampling. A total of 399 HIV-positive patients underwent upper gastrointestinal (UGI) endoscopy. HIV-positive patients with EU determined by UGI endoscopy followed by biopsies were analyzed by the hematoxylin-eosin (HE) and immunohistochemical (IH) methods. EU was detected in 41 patients (mean age, 39.2 years; 23 males), with a prevalence of 10.27%. The median CD4 count was 49 cells/mm(3) (range, 1-361 cells/mm(3)) and the viral load was 58,869 copies per milliliter (range, 50-77,3290 copies per milliliter). UGI endoscopy detected 29 of 41 EU suggestive of cytomegalovirus (CMV) infection and 7 of 41 indicating herpes simplex virus (HSV) infection. HE histology confirmed 4 of 29 ulcers induced by CMV, 2 of 7 induced by HSV, and 1 of 7 induced by HSV plus CMV. IH for CMV and HSV confirmed the HE findings and detected one additional CMV-induced case. UGI endoscopy showed 100% sensitivity and 15% specificity for the diagnosis of EU due to CMV or HSV compared to HE and IH. HE proved to be an adequate method for etiologic evaluation, with 87% sensitivity and 100% specificity compared to IH. The number of samples did not influence the etiologic evaluation. The data support the importance of IH as a complementary method for HE in the diagnosis of EU of viral etiology.