AnintroductiontoBayesianNetworksandtheBayesNet.pptVIP

An introduction to Bayesian Networksand the Bayes Net Toolboxfor Matlab Kevin Murphy MIT AI Lab 19 May 2003 Outline An introduction to Bayesian networks An overview of BNT What is a Bayes (belief) net? What is a Bayes net? Why are Bayes nets useful? - Graph structure supports - Modular representation of knowledge - Local, distributed algorithms for inference and learning - Intuitive (possibly causal) interpretation What can Bayes nets be used for? A real Bayes net: Alarm More real-world BN applications “Microsoft’s competitive advantage lies in its expertise in Bayesian networks”-- Bill Gates, quoted in LA Times, 1996 MS Answer Wizards, (printer) troubleshooters Medical diagnosis Genetic pedigree analysis Speech recognition (HMMs) Gene sequence/expression analysis Turbocodes (channel coding) Dealing with time In many systems, data arrives sequentially Dynamic Bayes nets (DBNs) can be used to model such time-series (sequence) data Special cases of DBNs include State-space models Hidden Markov models (HMMs) State-space model (SSM)/Linear Dynamical System (LDS) Example: LDS for 2D tracking Hidden Markov model (HMM) Probabilistic graphical models Toy example of a Markov net A real Markov net Inference Kalman filtering (recursive state estimation in an LDS) Forwards algorithm for HMMs Message passing view of forwards algorithm Forwards-backwards algorithm Belief Propagation BP: parallel, distributed version Representing potentials For discrete variables, potentials can be represented as multi-dimensional arrays (vectors for single node potentials) For jointly Gaussian variables, we can use y(X) = (m, S) or y(X) = (S-1 m ,S-1) In general, we can use mixtures of Gaussians or non-parametric forms Manipulating discrete potentials Manipulating Gaussian potentials Closed-form formulae for marginalization and multiplication O(1)/O(n3) complexity per operation Mixtures of Gaussian potentials are not closed under marginalization, so need approximations (moment mat