本章目录

19 Undirected graphical models (Markov random fields) 661
19.1 Introduction 661
19.2 Conditional independence properties of UGMs 661
19.2.1 Key properties 661
19.2.2 An undirected alternative to d-separation 663
19.2.3 Comparing directed and undirected graphical models 664
19.3 Parameterization of MRFs 665
19.3.1 The Hammersley-Clifford theorem 665
19.3.2 Representing potential functions 667
19.4 Examples of MRFs 668
19.4.1 Ising model 668
19.4.2 Hopfield networks 669
19.4.3 Potts model 671
19.4.4 Gaussian MRFs 672
19.4.5 Markov logic networks * 674
19.5 Learning 676
19.5.1 Training maxent models using gradient methods 676
19.5.2 Training partially observed maxent models 677
19.5.3 Approximate methods for computing the MLEs of MRFs 678
19.5.4 Pseudo likelihood 678
19.5.5 Stochastic maximum likelihood 679
19.5.6 Feature induction for maxent models * 680
19.5.7 Iterative proportional fitting (IPF) * 681
19.6 Conditional random fields (CRFs) 684
19.6.1 Chain-structured CRFs, MEMMs and the label-bias problem 684
19.6.2 Applications of CRFs 686
19.6.3 CRF training 692
19.7 Structural SVMs 693
19.7.1 SSVMs: a probabilistic view 693
19.7.2 SSVMs: a non-probabilistic view 695
19.7.3 Cutting plane methods for fitting SSVMs 698
19.7.4 Online algorithms for fitting SSVMs 700
19.7.5 Latent structural SVMs 701

github下载链接：https://github.com/916718212/Machine-Learning-A-Probabilistic-Perspective-.git