Lectures 丨 References
课件版权归作者所有。未经作者同意,禁止公开散布。
Slides 要点 Readings
lesson01.pdf 图模型入门,综述 KF book 1.2
lesson02_Semantics.pdf
    图模型语义
  • 有向(无环)概率图模型
  • 无向概率图模型
    Required:
  • CDLS book Chapter 5
  • Murphy book 10.1 DGMs
  • Murphy book 19.2 UGMs
    Optional:
  • Bishop book 8.1 - 8.3
lesson03_HMM_CRF.pdf 举例:隐马尔科夫模型(Hidden Markov Models, HMM),条件随机场(Conditional Random Fields, CRF)
    Required:
  • Rabiner's HMM tutorial 1989.
  • CRF: probabilistic models for segmenting and labeling sequeance data, ICML
  • H. Wallach, Conditional Random Fields: An Introduction, 2004.
  • TMG-THMM
    Optional:
  • Murphy book 17.3 Hidden Markov models (HMMs)
  • Murphy book 19.6 Conditional random fields (CRFs)
lesson04_mlEstimate.pdf 最大似然参数估计
    Required:
  • Murphy book 11.4 The EM algorithm
    Optional:
  • A. Dempster et al., EM algorithm original paper
  • Gu & Kong, SA 1998.
  • Delyon, Lavielle and Moulines, SAEM, 1999.
lesson05_RFLearning.pdf 无向概率图模型参数估计
    Required:
  • Murphy book 19.3 Parameterization of MRFs
  • Murphy book 19.5 Learning
    Optional:
  • Robbins and Monro (1951) SA
  • Inducing features of random fields 1997
  • Ruslan Salakhutdinov Ph.D. Thesis 2.1 Restricted Boltzmann Machines
  • Trans-dimensional Random Field Language Model ACL2015
  • TCRF-based Confidence Measures ICASSP2012
lesson06_ve.pdf 变量消除算法 — 图模型(精确)推理算法的早期形式,易于理解
    Required:
  • DECHTER, BUCKET ELIMINATION, 1996
  • Rabiner's HMM tutorial 1989.
    Optional:
  • Murphy book 20.3 The variable elimination algorithm
lesson07_ClusterTree.pdf 图模型(精确)推理算法的完整介绍
    Required:
  • Unifying Cluter-Tree Decomposition 2005.
  • Factor Graphs and the Sum-Product Algorithm, 2001
    Optional:
  • Probability Propagation, 1990.
  • Murphy book 20.4 The junction tree algorithm
lesson08_triangulate.pdf 三角化
    Required:
  • Murphy thesis 2002 B.3 From graph to junction tree
    Optional:
  • CDLS book 4.4 From chain graph to junction tree
lesson09_Kalman.pdf 含有连续变量的图模型的推理
    Required:
  • Murphy book 18.3 Inference in LG-SSM
  • Murphy thesis 2002 B.5 Message passing with continuous random variables
    Optional:
  • Murphy book 18.5 Approximate online inference for non-linear, non-Gaussian SSMs
lesson10_sampling.pdf 近似推理算法 — 采样近似
    Required:
  • Murphy book 23 Monte Carlo Inference
  • Murphy book 24 Markov Chain Monte Carlo (MCMC) Inference
    Optional:
  • MacKay, Introduction to Monte Carlo methods, 1999.
lesson11_variational.pdf 近似推理算法 — 变分近似
    Required:
  • Murphy book 21 Variational Inference
  • Murphy book 22 More Variational Inference
    Optional:
  • Jaakkola, Tutorial on variational approximation methods, 2000.
  • Jordan and Ghahramani, An Introduction to Variational Methods for GMs, 2001
lesson12_BayesEstimate.pdf 贝叶斯学习
    Required:
  • Murphy book 5 Bayesian Statistics
  • MacKay book 28 Model Comparison and Occam's Razor
    Optional:
  • Bishop book 2.Probability Distributions
  • Variational nonparametric Bayesian hidden Markov Model 2010.
  • Bayesian K-Means 2006.
lesson13_StructureLearning.pdf 结构学习
    Required:
  • Murphy book 26 Graphical Model Structure Learning
    Optional:
  • Heckerman, A tutorial on learning with BNs, 1995
  • Original papers on structural EM
lesson14_lda.pdf TopicModeling - 一个综合的例子,展示在实际中运用图模型的表示、推理和学习理论
    Required:
  • Blei, Ng and Jordan. Latent Dirichlet Allocation, NIPS 2002.
    Optional:
  • Blei, Ng and Jordan. Latent Dirichlet Allocation, JMLR 2003.
  • Topic-weak-correlated LDA 2010.