Sequential Monte Carlo Methods Homepage 

Credits and Disclaimers

UK Mirror 
Sequential learning and inference methods are important
in many applications involving realtime signal processing,
where data arrival is inherently sequential. Furthermore, one might wish
to adopt a sequential processing strategy to deal with nonstationarity in
signals, so that information from the recent past is given greater weighting
than information from the distant past. Computational simplicity in the
form of not having to store all the data might also constitute an additional motivating
factor for sequential methods.
Many realworld signal processing problems involve elements of nonGaussianity, nonlinearity and nonstationarity. Consequently, it is not usually possible to derive exact closed form estimators based upon the standard criteria of maximum likelihood, maximum a posteriori or minimum meansquared error. Classical suboptimal methods, such as extended Kalman filters and Gaussian sum approximations are easy to implement. However, they do not take into account all the salient statistical features of the processes under consideration, thereby often leading to poor results. The advent of cheap and massive computational power, in conjunction with some recent developments in applied statistics, have stimulated many advancements in the field of Sequential Monte Carlo simulation. Monte Carlo methods are very flexible in that they do not require any assumptions about the probability distributions of the data. Moreover, experimental evidence suggests that these methods lead to improved results. From a Bayesian perspective, Sequential Monte Carlo methods allow one to compute the posterior probability distributions of interest online. Yet, the methods can also be applied within a maximum likelihood context. As a result, they are being applied to a large number of interesting real problems such as computer vision, econometrics, medical prognosis, tracking, communications, blind deconvolution, statistical model diagnostic tools, automatic control, navigation and neural network training among others. Comments? Ideas? Do not hesitate to email: op205@eng.cam.ac.uk 