Search

Loudspeaker design requires accurate models of driver voice coil impedance. This paper examines three model classes (standard, Leach, and van Maanen) from the audio literature and compares them using Bayesian model comparison via nested sampling. Data is generated from impedance measurements of two commercial loudspeaker drivers. We conclude that, for most design tasks involving these drivers, the van Maanen model with 3 lossy inductance groups is the most appropriate model.

This paper presents a method for designing intermodal freight rail-to-highway terminals using Bayesian model selection. Each model consists of a unique network of cities and rail terminals. The model parameters are the amount of freight flowing on each individual path in the network. The prior distribution constrains the freight flowing on each path between each pair of origin and destination cities to Gaussian distributions with means corresponding to the estimated freight demand between the cities. The likelihood function favors low total cost in the network and penalizes routing schemes that exceed terminal capacities. Nested sampling is used to compute the Bayesian evidence for each model. The paper describes the generation of, results from, and conclusions drawn from a realistic test case.

Graphical representations of the independence relations in joint probability functions—commonly known as Bayesian networks—have played an essential role in knowledge representation and decision support. In this paper, we present a graphical representation of the independence relations in multiattribute utility functions, which we call bidirectional utility diagrams. These diagrams apply to more general functional forms that relax both the grounding conditions and n‐increasing conditions of joint cumulative probability functions. We also provide an expansion theorem for multiattribute utility functions and show how special cases of this expansion lead to some of the well‐known formulations of utility independence that occur in the literature.

In biomedical studies, detecting the changes in a response distribution under different testing conditions is one of the important issues. For example, increase in dose level may lead to increase or decrease in the gene expression level. To address this issue, we propose a nonparametric Bayesian test for testing the difference in location when samples are collected under two different conditions. We apply Dirichlet process priors to estimate the probabilities, which imply constraint on cumulative distribution functions of occurrence evaluated at cut‐off value that partitions the expression range of that gene into two intervals. The proposed test can be easily extended for multiple samples comparisons in gene expression analysis.

Modern data mining and bioinformatics have presented an important playground for statistical learning techniques, where the number of input variables is possibly much larger than the sample size of the training data. In supervised learning, logistic regression or probit regression can be used to model a binary output and form perceptron classification rules based on Bayesian inference. In these cases , there is a lot of interest in searching for sparse model in High Dimensional regression(/classification) setup. we first discuss two common challenges for analyzing high dimensional data. The first one is the curse of dimensionality. The complexity of many existing algorithms scale exponentially with the dimensionality of the space and by virtue of that algorithms soon become computationally intractable and therefore inapplicable in many real applications. secondly, multicollinearities among the predictors which severely slowdown the algorithm. In order to make Bayesian analysis operational in high dimensio...

Ensembles provide a powerful method for improving the performance of automated classifiers by constructing piecewise models that combine individual component classifier hypotheses. Together, the combined output of the component classifiers is more capable of fitting the type of complex decision boundaries in data sets where class boundaries overlap and class exemplars are disperse in feature space. A key ingredient to ensemble classifier induction is error diversity among component classifiers. Work in the ensemble literature suggests that ensemble construction should consider diversity even at some expense to individual classifier performance. To make such tradeoffs, a component classifier inducer requires knowledge of the choices made by its peers in the ensemble. In this work, we present a method called MaxEnt‐DiSCO that trains component classifiers collectively using entropy as a measure of error diversity. Using the maximum entropy framework, we share information on instance selection among component...

The Spatial Sensitivity Function (SSF) is used to quantify a detector’s sensitivity to a spatially‐distributed input signal. By weighting the incoming signal with the SSF and integrating, the overall scalar response of the detector can be estimated. This project focuses on estimating the SSF of a light intensity sensor consisting of a photodiode. This light sensor has been used previously in the Knuth Cyberphysics Laboratory on a robotic arm that performs its own experiments to locate a white circle in a dark field (Knuth et al., 2007). To use the light sensor to learn about its surroundings, the robot’s inference software must be able to model and predict the light sensor’s response to a hypothesized stimulus. Previous models of the light sensor treated it as a point sensor and ignored its spatial characteristics. Here we propose a parametric approach where the SSF is described by a mixture of Gaussians (MOG). By performing controlled calibration experiments with known stimulus inputs, we used nested sampling to estimate the SSF of the light sensor using an MOG model with the number of Gaussians ranging from one to five. By comparing the evidence computed for each MOG model, we found that one Gaussian is sufficient to describe the SSF to the accuracy we require. Future work will involve incorporating this more accurate SSF into the Bayesian machine learning software for the robotic system and studying how this detailed information about the properties of the light sensor will improve robot’s ability to learn.

The problem of attaining a usable wireless connection at an arbitrary location is one of great concern to mobile end users. The majority of antennae currently in use for mobile devices conducting two way communications are omnidirectional. The use of a directional antenna allows for increased effective coverage area without increasing power consumption. However, directional antennae must be oriented toward a wireless network access point in order for their benefits to be realized. This paper outlines a system for determining the optimal orientation of a directional antenna without the need for additional hardware. The response of the antenna is described by the use of a parameterized model corresponding to the sum of a set of cardioid functions. Signal strength is measured at several antenna orientations and is used by a Metropolis‐Hastings search algorithm to estimate the model parameter values that best describe the antenna’s response pattern. Using this model the antenna can be oriented to respond optimally to the wireless network access point’s broadcast pattern.

The brains of premature infants must undergo significant maturation outside of the womb and are thus particularly susceptible to injury. Electroencephalographic (EEG) recordings are an important diagnostic tool in determining if a newborn’s brain is functioning normally or if injury has occurred. However, interpreting the recordings is difficult and requires the skills of a trained electroencephelographer. Because these EEG specialists are rare, an automated interpretation of newborn EEG recordings would increase access to an important diagnostic tool for physicians. To automate this procedure, we employ Bayesian probability theory to compute the posterior probability for the EEG features of interest and use the results in a program designed to mimic EEG specialists. Specifically, we will be identifying waveforms of varying frequency and amplitude, as well as periods of flat recordings where brain activity is minimal.

Cancellous bone is a lattice‐like arrangement of solid trabeculae surrounded by soft bone marrow. When interrogated with ultrasonic waves, the complex architecture gives rise to two longitudinal modes known as a fast and a slow wave. Depending on experimental conditions and the ultrasonic characteristics of the bone sample under investigation, the two waves may strongly overlap in the ultrasounic data. Analyzing such data conventionally, as if only one wave were present, can potentially mask or alter bone quality parameters that are commonly used in clinical sonometry.In this study, ultrasonic data were acquired on a bovine femur condyle specimen and on a plastic bone‐mimicking phantom constructed from Lucite and Lexan blocks. The acquired data were used as inputs to a program that implements a Bayesian calculation to model the ultrasound signal as two interfering plane waves and then estimates the ultrasonic parameters of the fast and slow waves. The calculations were carried out using Markov chain Monte Carlo (MCMC) with simulated annealing to approximate the Bayesian calculations.The models showed good agreement with the acquired data from both the bone and bone‐mimicking phantom. The sound velocities that maximized the joint posterior probability for the model of the bone‐mimicking fantom were 2765 m/s (fast wave) and 2193 m/s (slow wave), values that are moderately close to the true values for Lucite (2734 m/s) and Lexan (2185 m/s). These parameter estimates provide more reliable estimates of the material properties of the plastic bone‐mimicking phantom than conventional analysis. The mean ± standard deviation velocities in the bone sample at a representative site were 2036±4 m/s (fast wave) and 1511±1 m/s (slow wave).

The problem of automatic music description is considered. The recorded music is modeled as a superposition of known sounds from a library weighted by unknown weights. Similar observation models are commonly used in statistics and machine learning. Many methods for estimation of the weights are available. These methods differ in the assumptions imposed on the weights. In Bayesian paradigm, these assumptions are typically expressed in the form of prior probability density function (pdf) on the weights. In this paper, commonly used assumptions about music signal are summarized and complemented by a new assumption. These assumptions are translated into pdfs and combined into a single prior density using combination of pdfs. Validity of the model is tested in simulation using synthetic data.

Contemporary approaches to speech and speaker recognition decompose the problem into four components: feature extraction, acoustic modeling, language modeling and search. Statistical signal processing is an integral part of each of these components, and Bayes Rule is used to merge these components into a single optimal choice. Acoustic models typically use hidden Markov models based on Gaussian mixture models for state output probabilities. This popular approach suffers from an inherent assumption of linearity in speech signal dynamics. Language models often employ a variety of maximum entropy techniques, but can employ many of the same statistical techniques used for acoustic models.In this paper, we focus on introducing nonlinear statistical models to the feature extraction and acoustic modeling problems as a first step towards speech and speaker recognition systems based on notions of chaos and strange attractors. Our goal in this work is to improve the generalization and robustness properties of a spe...

Over the last few years there has been an ongoing effort at the Biomedical Magnetic Resonance Laboratory within Washington University to develop data analysis applications using Bayesian probability theory. A few of these applications are specific to Magnetic Resonance data, however, most are general and can analyze data from a wide variety of sources. These data analysis applications are server based and they have been written in such a way as to allow them to utilize as many processors as are available. The interface to these Bayesian applications is a client based Java interface. The client, usually a Windows PC, runs the interface, sets up an analysis, sends the analysis to the server, fetches the results and displays the appropriate plots on the users client machine. Together, the client and server software can be used to solve a host of interesting problems that occur regularly in the sciences. In this paper, we describe both the client and server software and briefly discuss how to acquire, install...

In the automotive industry, the necessary reduction of pollutant emission for new Diesel engines requires the control of combustion events. This control is efficient provided combustion parameters such as combustion occurrence and combustion energy are relevant. Combustion parameters are traditionally measured from cylinder pressure sensors. However this kind of sensor is expensive and has a limited lifetime. Thus this paper proposes to use only one cylinder pressure on a multi‐cylinder engine and to extract combustion parameters from the other cylinders with low cost knock sensors. Knock sensors measure the vibration circulating on the engine block, hence they do not all contain the information on the combustion processes, but they are also contaminated by other mechanical noises that disorder the signal. The question is how to combine the information coming from one cylinder pressure and knock sensors to obtain the most relevant combustion parameters in all engine cylinders. In this paper, the issue is ...

Functional magnetic resonance imaging (fMRI) is an important technique for neuroimaging. The conventional system identification methods used in fMRI data analysis model the whole brain as a stationary linear system characterized by its impulse response defined by the hemodynamic response function (HFR). Modeling and estimating the HFR in fMRI experiments is an important aspect of the analysis of functional neuroimages. This work uses a Bayesian approach to linear systems analysis with Gaussian noise to make inferences about the HRF. A new method based on maximum entropy considerations is proposed to define the prior involved in estimating the posterior distribution characterizing the HRF. Using the knowledge acquired regarding the HRF, an hypothesis test can easily be derived to infer activation detection in individual pixels. The performances of the proposed method are evaluated in an application example.

A statistical test for detecting activated pixels in functional MRI (fMRI) data is proposed. For the derivation of this test, the fMRI time series measured at each voxel is modeled as the sum of a response signal which arises due to the experimentally controlled activation‐baseline pattern, a nuisance component representing effects of no interest, and Gaussian white noise. The test is based on comparing the dimension of the voxels fMRI time series fitted data models with and without controlled activation‐baseline pattern. The Jeffrey divergence is used for this comparison. The test has the advantage of not requiring a level of significance or a threshold to be provided.

We construct and apply a stochastic Gaussian Process (GP) model of flight trajectory generation for pigeons trained to home from specific release sites. The model shows increasing predictive power as the birds become familiar with the sites, mirroring the animal's learning process. We show how the increasing similarity between successive flight trajectories can be used to infer, with increasing accuracy, an idealised route that captures the repeated spatial aspects of the bird's flight. We subsequently use techniques associated with reduced-rank GP approximations to objectively identify the key waypoints used by each bird to memorise its idiosyncratic habitual route between the release site and the home loft. © 2009 American Institute of Physics.

A surface impedance function (SIF) appropriate for use on the aperture surface of a conducting body with a dielectric-filled cavity, is presented. Unlike the usual SIFs that might be used on an aperture, this SIF takes into account not only the wave transmitted through the aperture but also the wave reflected from the inside of the cavity the shape of the aperture and cavity, and the polarization and direction of the incident wave. The SIF is derived heuristically from the series-reflection solution for a plane wave normally incident on an infinite flat conducting plate with a flat dielectric coating. The SIF was developed and used in a combined method of moments solution for the scattered fields due to an incident plane wave. This combined technique greatly reduces the number of current expansion coefficients to be determined using the method of moments and hence also reduces the number of impedance elements required for calculation in the method of moments. Application of the SIF in a combined method is illustrated for a two-dimensional object. >

Transverse electric (TE) and transverse magnetic (TM) scattering from dielectric-filled, cavity-backed apertures in two-dimensional bodies are treated using the method of moments technique to solve a set of combined-field integral equations for the equivalent induced electric and magnetic currents on the exterior of the scattering body and on the associated aperture. Results are presented for the backscatter radar cross section (RCS) versus the electrical size of the scatterer for two different dielectric-filled cavity-backed geometries. The first geometry is a circular cylinder of infinite length which has an infinite length slot aperture along one side. The cavity inside the cylinder is dielectric filled and is also of circular cross section. The two cylinders (external and internal) are of different radii and their respective longitudinal axes are parallel but not collocated. The second is a square cylinder of infinite length which has an infinite length slot aperture along one side. The cavity inside the square cylinder is dielectric-filled and is also of square cross section. >

This paper describes the use of the Bayesian inference framework for the design of linear antenna arrays. The principal advantage of using the Bayesian inference framework for array design is that it makes possible the automatic determination of the number of radiators required to meet given design requirements. The inference framework achieves this by making accessible to the array designer powerful computational tools developed for the simultaneous solution of parameter estimation and model selection problems.

The problem of determining the scattering cross section of an arbitrarily shaped two-dimensional conducting body with an arbitrarily shaped dielectric filled cavity is considered. The problem is solved using a method-of-moments solution for the combined field integral equations. The particular form of the method of moments solution used here uses a minimum number of expansion coefficients. Results are given for transverse electric and transverse magnetic incident waves. >

Electromagnetic induction (EMI) landmine detection can be cast as a Bayesian model comparison problem. The models used for low metallic‐content mine detection are based on the equivalent electrical circuit representation of the EMI detection system. The EMI detection system is characterized and modeled by the pulse response of its equivalent circuit. The analytically derived transfer function between the transmitter coil and receiver coil demonstrates that the EMI detection system is a third order system in the absence of a mine and that the presence of a mine adds an additional pole that makes the detection system fourth order. The value of the additional pole is determined by the equivalent inductance and resistance of the mine and is unique for each mine type. This change in system order suggests that measured system pulse responses can be used in conjunction with pulse response models to infer the presence or absence of a landmine. The difficulty of this technique is that the amplitude of the term added to the the system pulse response by the landmine is small compared to the pulse response of the system alone. To test the feasibility of Bayesian inference based EMI landmine detection, an EMI detection system experiment was designed and built. In the experiment the EMI detection system was driven by a broadband maximal‐length sequence (MLS) in order to obtain sufficient dynamic range in the measured pulse responses. This paper presents the parameterized pulse response models for the detection system with and without a landmine present and gives appropriate priors for the parameters of these models. This paper also presents the ratios of computed posterior probabilities for the mine and no mine models based on data obtained from the experimental EMI landmine detection system. These odds demonstrate the potential for Bayesian EMI landmine detection.

This paper gives an algorithm for calculating posterior probabilities using thermodynamic integration. The thermodynamic integration calculations are accomplished by annealing an ensemble of Markov chains with an adaptive schedule. The algorithm includes a method for determining “good” starting positions for the chains at each new value of the annealing parameter.

To demonstrate the calculation of scattered fields from bodies with thin features, near-self impedance element, calculation has been incorporated into the two-dimensional MM (moment method) program of P.M. Goggans and T.H. Shumpert (1990). This MM program solves the combined-field integral-equation for composite bodies using linear zones with pulse current expansion functions and point matching (delta function testing). The bistatic RCS (radar cross section) of a dielectric-coated conducting cylinder for a TE-polarized incident wave is plotted. The calculated RCS is plotted with and without the use of near-self impedance elements. Near exact agreement with the series solution is obtained by use of the near-self impedance elements. >

Landmine detection can be cast as a model selection problem in which probability theory is used as logic for inductive inference. Using this method, the landmine detection decision is based on the values of calculated posterior probabilities for two propositions: 'The received signal is from a landmine' and 'The received signal is from the background.' The posterior probability for a proposition is the probability for the proposition given the observed data signal and the information known prior to the observation. Calculation of the posterior probability requires the numerical integration of a multi-dimensional probability density function. Until the beginning of the last decade, there were few robust methods available to perform these numeral integrations and no methods that could be generally applied. As a result, probability theory as logic for inductive inference found only infrequent use in practical detection algorithms. Because of the increasing power of computers and new research in the areas of Markov chain Monte Carlo and multi-dimensional adaptive-quadrature integration methods, practical detection algorithms based on the use of probability theory as logic for inductive inference are now being developed and used. This paper describes our model selection formulation of the landmine detection problem and presents results obtained using multi-dimensional adaptive quadrature.

This paper discusses the technology, data and ground surface-velocity models to be used in addressing the identification/detection inference problem engendered by a recently developed acoustic landmine detection system. The landmine detection problem is formulated as a model selection problem and important aspects of computing the evidence for the models are given.

Probability as logic is used to estimate the surface velocity of a patch of soil driven by an incident acoustic wave. The data used by the estimation procedure is obtained from a laser Doppler vibrometer (LDV). The output of the LDV is an intermediate-frequency carrier that is frequency- modulated by the soil surface velocity. Additionally, the LDV output is amplitude modulated by an undesirable variation in the returned laser signal due to dynamic optical speckle. The effect of the amplitude modulated by an undesirable variation in the returned laser signal due to dynamical optical speckle. The effect of the amplitude modulation on the estimate of the soil surface velocity is illustrated with results obtained using the Markov chain Monte Carlo method.

Sound waves from a powerful loudspeaker can excite a certain type of vibration of the surface of the ground whena mine (or other firm, smooth object) is present and near the surface. In turn, a laser-Doppler vibrometer can beemployed to acquire information about the surface vibrations. In particular, the portion of the ground surface thatis vibrating has the shape of the projection of the mine onto the surface. This paper discusses a method basedon Bayesian probability theory for processing laser-Doppler vibrometer data to infer the periphery of any surfacevibration pattern. Difficulties with using a phase-lock loop in determining a surface map are also discussed.Keywords: Mine detection, Bayesian parameter estimation, laser-Doppler vibrometer 1. INTRODUCTION Buried mines pose a grave danger to inhabitants of many parts of the world; additionally, large tracts of potentially valuable land have been rendered practically useless by all sorts of unexploded ordnance. Finding and destroying these

An attempt to accurately analyze hard struts without using surface impedance models is presented. The authors consider initially only a strut coated with a dielectric layer, i.e., with no corrugations and no metallic strips on it. This strut coated with a dielectric layer will have minimum forward scattering, for the TM (transverse magnetic)-case only. It does not retain its minimum scattering characteristics for the TE (transverse electric)-case, for which the corrugations or strips are needed. Still, such a strut is applicable in a linearly polarized system, to reduce scattering for TM-case blockage losses, e.g., support struts located in the E-plane of axisymmetric reflector antennas. The calculation model is tested against measured results. >