Gaussian linear modelling cannot address current signal processing demands. Inmodern contexts, such as Independent Component Analysis (ICA), progress has beenmade specifically by imposing non-Gaussian and/or non-linear assumptions. Hence,standard Wiener and Kalman theories no longer enjoy their traditional hegemony inthe field, revealing the standard computational engines for these problems. In theirplace, diverse principles have been explored, leading to a consequent diversity in theimplied computational algorithms. The traditional on-line and data-intensive preoccupationsof signal processing continue to demand that these algorithms be tractable.Increasingly, full probability modelling (the so-called Bayesian approach)—orpartial probability modelling using the likelihood function—is the pathway for designof these algorithms. However, the results are often intractable, and so the areaof distributional approximation is of increasing relevance in signal processing. TheExpectation-Maximization (EM) algorithm and Laplace approximation, for example,are standard approaches to handling difficult models, but these approximations(certainty equivalence, and Gaussian, respectively) are often too drastic to handlethe high-dimensional, multi-modal and/or strongly correlated problems that are encountered.Since the 1990s, stochastic simulation methods have come to dominateBayesian signal processing. Markov Chain Monte Carlo (MCMC) sampling, and relatedmethods, are appreciated for their ability to simulate possibly high-dimensionaldistributions to arbitrary levels of accuracy. More recently, the particle filtering approachhas addressed on-line stochastic simulation. Nevertheless, the wider acceptabilityof these methods—and, to some extent, Bayesian signal processing itself—has been undermined by the large computational demands they typically make.The Variational Bayes (VB) method of distributional approximation originates—as does the MCMC method—in statistical physics, in the area known as Mean FieldTheory. Its method of approximation is easy to understand: conditional independenceis enforced as a functional constraint in the approximating distribution, andthe best such approximation is found by minimization of a Kullback-Leibler divergence(KLD). The exact—but intractable—multivariate distribution is therefore factorizedinto a product of tractable marginal distributions, the so-called VB-marginals.This straightforward proposal for approximating a distribution enjoys certain optimality properties. What is of more pragmatic concern to the signal processing community,however, is that the VB-approximation conveniently addresses the followingkey tasks:1. The inference is focused (or, more formally, marginalized) onto selected subsetsof parameters of interest in the model: this one-shot (i.e. off-line) use of the VBmethod can replace numerically intensive marginalization strategies based, forexample, on stochastic sampling.2. Parameter inferences can be arranged to have an invariant functional formwhen updated in the light of incoming data: this leads to feasible on-linetracking algorithms involving the update of fixed- and finite-dimensional statistics.In the language of the Bayesian, conjugacy can be achieved under theVB-approximation. There is no reliance on propagating certainty equivalents,stochastically-generated particles, etc.Unusually for a modern Bayesian approach, then, no stochastic sampling is requiredfor the VB method. In its place, the shaping parameters of the VB-marginals arefound by iterating a set of implicit equations to convergence. This Iterative VariationalBayes (IVB) algorithm enjoys a decisive advantage over the EM algorithmwhose computational flow is similar: by design, the VB method yields distributionsin place of the point estimates emerging from the EM algorithm. Hence, in commonwith all Bayesian approaches, the VB method provides, for example, measures ofuncertainty for any point estimates of interest, inferences of model order/rank, etc.The machine learning community has led the way in exploiting the VB methodin model-based inference, notably in inference for graphical models. It is timely,however, to examine the VB method in the context of signal processing where, todate, little work has been reported. In this book, at all times, we are concerned withthe way in which the VB method can lead to the design of tractable computationalschemes for tasks such as (i) dimensionality reduction, (ii) factor analysis for medicalimagery, (iii) on-line filtering of outliers and other non-Gaussian noise processes, (iv)tracking of non-stationary processes, etc. Our aim in presenting these VB algorithmsis not just to reveal new flows-of-control for these problems, but—perhaps moresignificantly—to understand the strengths and weaknesses of the VB-approximationin model-based signal processing. In this way, we hope to dismantle the current psychologyof dependence in the Bayesian signal processing community on stochasticsampling methods.Without doubt, the ability to model complex problems to arbitrarylevels of accuracy will ensure that stochastic sampling methods—such as MCMC—will remain the golden standard for distributional approximation. Notwithstandingthis, our purpose here is to show that the VB method of approximation can yieldhighly effective Bayesian inference algorithms at low computational cost. In showingthis, we hope that Bayesian methods might become accessible to a much broaderconstituency than has been achieved to date。
猜您喜欢
推荐内容
开源项目推荐 更多
热门活动
热门器件
用户搜过
随便看看
热门下载
评论