Tài liệu hạn chế xem trước, để xem đầy đủ mời bạn chọn Tải xuống
1
/ 144 trang
THÔNG TIN TÀI LIỆU
Thông tin cơ bản
Định dạng
Số trang
144
Dung lượng
0,99 MB
Nội dung
[...]... Weigend (2:1991), and Buntine (2:1992) have done interesting work usingmethods other than MonteCarlo I have applied Markov chain MonteCarlo methods to some of the same problems (Neal, 2:1992a, 2:1992c, 2:1993a) Though these applications to problems in articial intelligence are still in their infancy, I believe the Markov chain MonteCarlo approach has great potential as a widely applicable computational... precisely the class of problems for which use of MonteCarlomethods based on Markov chains is appropriate, and discuss why these problems are dicult to solve by other methods I also present the basics of the theory of Markov chains, and discuss recently developed theoretical techniques that may allow useful analytical results to be derived for the complex chains encountered in practical work Section... framework Outline of this review In Section 2, which follows, I discuss probabilisticinference and its applications in articial intelligence This topic can be divided into inferenceusing a specied model, and statistical inference concerning the model itself In both areas, I indicate where computational problems arise for which MonteCarlomethods may be appropriate I also present some basic concepts of... realizations of this basic concept that have been developed, and to relate these methods to problems of probabilistic reasoning and empirical learning in articial intelligence I will be particularly concerned with the potential for Markov chain MonteCarlo methods to provide computationally feasible implementations of Bayesian inference and learning In my view, the Bayesian approach provides a exible framework... approach provides a exible framework for representing the intricate nature of the world and our knowledge of it, and the MonteCarlomethods I will discuss provide a correspondingly exible mechanism for inference within this framework Historical development Sampling methods based on Markov chains were rst developed for applications in statistical physics Two threads of development were begun forty years... on methods applicable to expert systems and other high-level reasoning tasks, and of Szelisksi (2:1989), on low2 1 Introduction level vision Much of the recent work on \neural networks", such as that described by Rumelhart, McClelland, and the PDP Research Group (2:1986), can also be regarded as statistical inference for probabilistic models Applications in articial intelligence of Markovchain Monte. .. X as t increases, and so that the Markovchain can feasibly be simulated by sampling from the initial distribution and then, in succession, from the conditional transition distributions For a suciently long chain, equation (1.2) can then be used to estimate expectations 1 1 Introduction Typically, the Markovchain explores the space in a \local" fashion In some methods, for example, x(t) diers from... all the algorithms discussed I conclude in Section 7 by discussing possibilities for future research concerning Markov chain MonteCarlo algorithms and their applications Finally, I have included a comprehensive, though hardly exhaustive, bibliography of work in the area 3 2 ProbabilisticInference for Articial Intelligence Probability and statistics provide a basis for addressing two crucial problems... Theory of Markov chains : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 36 4.1 4.2 4.3 4.4 Gibbs sampling : : : : : : : : : : : : : : : : : : : : : : : : : The Metropolis algorithm : : : : : : : : : : : : : : : : : : : Variations on the Metropolis algorithm : : : : : : : : : : : : Analysis of the Metropolis and Gibbs sampling algorithms : 5 The Dynamical and Hybrid Monte Carlo Methods 5.1... it may dier with respect to xi , for some i, but have x(t) = x(t 1) for j 6= i Other methods j j may change all components at once, but usually by only a small amount Locality is often crucial to the feasibility of these methods In the Markovchain framework, it is possible to guarantee that such step-by-step local methods eventually produce a sample of points from the globally-correct distribution