Increasingly, for many application areas, it is becoming important to include elements of nonlinearity and non-Gaussianity in order to model accurately the underlying dynamics of a physical system. Moreover, it is typically crucial to process data on-line as it arrives, both from the point of view of storage costs as well as for rapid adaptation to changing signal characteristics. In this paper, we review both optimal and suboptimal Bayesian algorithms for nonlinear/non-Gaussian tracking problems, with a focus on particle filters. Particle filters are sequential Monte Carlo methods based on point mass (or "particle") representations of probability densities, which can be applied to any state-space model and which generalize the traditional Kalman filtering methods. Several variants of the particle filter such as SIR, ASIR, and RPF are introduced within a generic framework of the sequential importance sampling (SIS) algorithm. These are discussed and compared with the standard EKF through an illustrative example.

/pdf/a-tutorial-on-particle-filters-for-online-nonlinear-non-1rzkwgk2pp.pdf

A tutorial on particle filters for online nonlinear/non-Gaussian Bayesian tracking

Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

Today's Web-enabled deluge of electronic data calls for automated methods of data analysis. Machine learning provides these, developing methods that can automatically detect patterns in data and then use the uncovered patterns to predict future data. This textbook offers a comprehensive and self-contained introduction to the field of machine learning, based on a unified, probabilistic approach. The coverage combines breadth and depth, offering necessary background material on such topics as probability, optimization, and linear algebra as well as discussion of recent developments in the field, including conditional random fields, L1 regularization, and deep learning. The book is written in an informal, accessible style, complete with pseudo-code for the most important algorithms. All topics are copiously illustrated with color images and worked examples drawn from such application domains as biology, text processing, computer vision, and robotics. Rather than providing a cookbook of different heuristic methods, the book stresses a principled model-based approach, often using the language of graphical models to specify models in a concise and intuitive way. Almost all the models described have been implemented in a MATLAB software package--PMTK (probabilistic modeling toolkit)--that is freely available online. The book is suitable for upper-level undergraduates with an introductory-level college math background and beginning graduate students.

Machine Learning : A Probabilistic Perspective

Monte Carlo methods are revolutionizing the on-line analysis of data in fields as diverse as financial modeling, target tracking and computer vision. These methods, appearing under the names of bootstrap filters, condensation, optimal Monte Carlo filters, particle filters and survival of the fittest, have made it possible to solve numerically many complex, non-standard problems that were previously intractable. This book presents the first comprehensive treatment of these techniques, including convergence results and applications to tracking, guidance, automated target recognition, aircraft navigation, robot navigation, econometrics, financial modeling, neural networks, optimal control, optimal filtering, communications, reinforcement learning, signal enhancement, model averaging and selection, computer vision, semiconductor design, population biology, dynamic Bayesian networks, and time series analysis. This will be of great value to students, researchers and practitioners, who have some basic knowledge of probability. Arnaud Doucet received the Ph. D. degree from the University of Paris-XI Orsay in 1997. From 1998 to 2000, he conducted research at the Signal Processing Group of Cambridge University, UK. He is currently an assistant professor at the Department of Electrical Engineering of Melbourne University, Australia. His research interests include Bayesian statistics, dynamic models and Monte Carlo methods. Nando de Freitas obtained a Ph.D. degree in information engineering from Cambridge University in 1999. He is presently a research associate with the artificial intelligence group of the University of California at Berkeley. His main research interests are in Bayesian statistics and the application of on-line and batch Monte Carlo methods to machine learning. Neil Gordon obtained a Ph.D. in Statistics from Imperial College, University of London in 1993. He is with the Pattern and Information Processing group at the Defence Evaluation and Research Agency in the United Kingdom. His research interests are in time series, statistical data analysis, and pattern recognition with a particular emphasis on target tracking and missile guidance.

/pdf/sequential-monte-carlo-methods-in-practice-3kfxeebn8e.pdf

Sequential Monte Carlo methods in practice

Planning and navigation algorithms exploit statistics gleaned from uncertain, imperfect real-world environments to guide robots toward their goals and around obstacles.

Probabilistic Robotics

Many tracking and guidance problems may be formulated as a terminating stochastic game in which the distribution of outcomes is affected by the intermediate actions. Traditional technique ignore this interaction. In this paper we develop an information gathering strategy which maximizes the expected gain of the outcome. For example, the objective could be a function of the terminal miss distance and target identify with penalties for missing a valid target or attacking a friendly one. Several trade-offs are addressed: the increased information available from taking more measurements, the fact that an increased number of measurement may adversely affect change of success and the fact that later measurements may be more informative but also may be of little use since there my not be enough time available for reaction to this extra information. The problem is formulated so that we are required to choose, under uncertainty, an alternative from a set of possible decisions. This set has a discrete uncertainty as to the number of measurements to be taken and a continuous uncertainty as to where and when the measurements should be taken. Preferences over consequences are modeled with a utility function. We propose to choose as optimal the alternative which maximizes expected utility. A simulation based approximation to the solution of this stochastic optimization problem is outlined. This relies on recent developments in dimensions swapping Markov Chain Monte Carlo (MCMC) techniques. The use of MCMC methodology allow us to explore the expected utility surface and thus select a measurement strategy. The resulting algorithm is demonstrated on a simple guidance problem.© (1998) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Bayesian sensor resource allocation

Summary Questionnaires were sent to 76 schools in the Macclesfield district. Seventy-four of these were returned by head teachers and/or reception teachers. Replies concerned the school entry medical examination, information on individual children, general health information and access to professionals. Additional comments on any aspects of the service were also solicited. The results have been analysed. The majority of teachers considered that the school entry medical examination is of value, but the quality of information reaching teachers is viewed more critically. Meetings were the most favoured means of communication, but there are obvious defects in the system. Some teachers did not even know of access channels. If improvements do not occur some of the value of the examination will be lost. Suggestions for better communications are made. With local management of schools, teacher opinion must be considered if a better school health service is to be provided for the children of the district.

The school entry medical examination. What do teachers think of it

A dynamic infrared scene projector based on IR luminescent devices has many potential advantages compared with existing systems based on micro-resistor arrays. These include very fast response times, as individual devices can be driven at frequencies greater than 1 MHz, and no need for cryogenic cooling. Additionally, luminescent sources can not only appear hot to an IR observer when in forward bias, but also appear cold in reverse bias (commonly referred to as negative luminescence), so that a large apparent temperature range around ambient can be simulated. For a scene projector a large array of photodiodes is required, where each photodiode can be biased individually. As a precursor to the manufacture of a scene projector, we have already fabricated large area MW devices, consisting of arrays of photodiodes, suitable for use as calibration sources in IR cameras. To reduce the currents needed to achieve maximum dynamic temperature range, we have used a novel micromachining technique to fabricate integrated optical concentrators in InSb/InAlSb devices. We present here recent results from a large area (~0.86cm2) medium wavelength (MW) device, consisting of an array of photodiodes each with an integrated optical concentrator. The reverse saturation current of the device was measured to be ~2.3A/cm2, which is significantly smaller than the value of ~9A/cm2 reported previously for similar devices without optical concentrators. The device also displays a large apparent temperature range in line with device modelling. Finally, we will discuss the perspectives on using similar devices for dynamic infrared scene projection.© (2003) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Perspectives on dynamic infrared scene projection using positive and negative luminescence

There is an increasingly important requirement for day and night, wide field of view imaging and tracking for both
imaging and sensing applications. Applications include military, security and remote sensing. We describe the
development of a proof of concept demonstrator of an adaptive coded-aperture imager operating in the mid-wave infrared
to address these requirements. This consists of a coded-aperture mask, a set of optics and a 4k x 4k focal plane array
(FPA). This system can produce images with a resolution better than that achieved by the detector pixel itself (i.e. superresolution)
by combining multiple frames of data recorded with different coded-aperture mask patterns. This superresolution
capability has been demonstrated both in the laboratory and in imaging of real-world scenes, the highest
resolution achieved being ½ the FPA pixel pitch. The resolution for this configuration is currently limited by vibration
and theoretically ¼ pixel pitch should be possible. Comparisons have been made between conventional and ACAI
solutions to these requirements and show significant advantages in size, weight and cost for the ACAI approach.

An adaptive coded aperture imager: building, testing and trialing a super-resolving terrestrial demonstrator

We describe uncooled mid-IR light emitting and negative luminescent diodes made form indium antimonide based III-V compounds, and long wavelength devices made from mercury cadmium telluride. The application of these devices to gas sensing, improved thermal imagers and imager testing is discussed.

Neil Gordon

Papers

Bayesian sensor resource allocation

The school entry medical examination. What do teachers think of it

Perspectives on dynamic infrared scene projection using positive and negative luminescence

An adaptive coded aperture imager: building, testing and trialing a super-resolving terrestrial demonstrator

4- to 10-um positive and negative luminescent diodes