Hidden semi-markov models : theory, algorithms and applications

Hidden semi-markov models theory, algorithms and applications

Hidden semi-Markov models (HSMMs) are among the most important models in the area of artificial intelligence / machine learning. Since the first HSMM was introduced in 1980 for machine recognition of speech, three other HSMMs have been proposed, with various definitions of duration and observation d...

Descripción completa

Detalles Bibliográficos
Otros Autores: Yu, Shun-Zheng, author (author)
Formato: Libro electrónico
Idioma:Inglés
Publicado: Amsterdam, [Netherlands] : Elsevier 2016.
Edición:1st edition
Colección:Computer science reviews and trends.
Materias:
Markov processes.
Renewal theory.
Ver en Biblioteca Universitat Ramon Llull:https://discovery.url.edu/permalink/34CSUC_URL/1im36ta/alma991009629829906719
Tabla de Contenidos:
  • Front Cover
  • Hidden Semi-Markov Models
  • Copyright Page
  • Contents
  • Preface
  • Acknowledgments
  • 1 Introduction
  • 1.1 Markov Renewal Process and Semi-Markov Process
  • 1.1.1 Markov Renewal Process
  • 1.1.2 Semi-Markov Process
  • 1.1.3 Generalized Semi-Markov Process
  • 1.1.4 Discrete-Time Semi-Markov Process
  • 1.2 Hidden Markov Models
  • 1.3 Dynamic Bayesian Networks
  • 1.4 Conditional Random Fields
  • 1.5 Hidden Semi-Markov Models
  • 1.6 History of Hidden Semi-Markov Models
  • 2 General Hidden Semi-Markov Model
  • 2.1 A General Definition of HSMM
  • 2.2 Forward-Backward Algorithm for HSMM
  • 2.2.1 Symmetric Form of the Forward-Backward Algorithm
  • 2.2.2 Initial Conditions
  • 2.2.3 Probabilities
  • 2.2.4 Expectations
  • 2.2.5 MAP Estimation of States
  • 2.3 Matrix Expression of the Forward-Backward Algorithm
  • 2.4 Forward-Only Algorithm for HSMM
  • 2.4.1 A General Forward-Only Algorithm
  • 2.4.2 Computing Smoothed Probabilities and Expectations
  • 2.5 Viterbi Algorithm for HSMM
  • 2.5.1 Bidirectional Viterbi HSMM Algorithm
  • 2.6 Constrained-Path Algorithm for HSMM
  • 3 Parameter Estimation of General HSMM
  • 3.1 EM Algorithm and Maximum-Likelihood Estimation
  • 3.1.1 EM Algorithm
  • 3.1.2 Derivation of Re-estimation Formulas
  • 3.2 Re-estimation Algorithms of Model Parameters
  • 3.2.1 Re-Estimation Algorithm for the General HSMM
  • 3.2.2 Supervised/Semi-Supervised Learning
  • 3.2.3 Multiple Observation Sequences
  • 3.3 Order Estimation of HSMM
  • 3.4 Online Update of Model Parameters
  • 3.4.1 Online Update Using Forward-Only Algorithm
  • 3.4.2 Online Update by Maximizing Likelihood Function
  • 3.4.3 Online Update for ML Segmentation
  • 4 Implementation of HSMM Algorithms
  • 4.1 Heuristic Scaling
  • 4.2 Posterior Notation
  • 4.3 Logarithmic Form
  • 4.4 Practical Issues in Implementation
  • 4.4.1 Nonindexable Observables.
  • 4.4.2 Missing Observables
  • 4.4.3 Unknown Model Order
  • 4.4.4 Unknown Observation Distribution
  • 4.4.5 Unknown Duration Distribution
  • 4.4.6 Unordered States
  • 4.4.7 Termination Condition for the Estimation Procedure
  • 5 Conventional HSMMs
  • 5.1 Explicit Duration HSMM
  • 5.1.1 Smoothed Probabilities
  • 5.1.2 Computational Complexity
  • 5.2 Variable Transition HSMM
  • 5.2.1 Smoothed Probabilities
  • 5.2.2 Computational Complexity
  • 5.3 Variable-Transition and Explicit-Duration Combined HSMM
  • 5.4 Residual Time HSMM
  • 5.4.1 Smoothed Probabilities
  • 5.4.2 Computational Complexity
  • 6 Various Duration Distributions
  • 6.1 Exponential Family Distribution of Duration
  • 6.2 Discrete Coxian Distribution of Duration
  • 6.3 Duration Distributions for Viterbi HSMM Algorithms
  • 7 Various Observation Distributions
  • 7.1 Typical Parametric Distributions of Observations
  • 7.2 A Mixture of Distributions of Observations
  • 7.2.1 Countable Mixture of Distributions
  • 7.2.2 Uncountable Mixture of Distributions
  • 7.3 Multispace Probability Distributions
  • 7.4 Segmental Model
  • 7.5 Event Sequence Model
  • 8 Variants of HSMMs
  • 8.1 Switching HSMM
  • 8.2 Adaptive Factor HSMM
  • 8.3 Context-Dependent HSMM
  • 8.4 Multichannel HSMM
  • 8.5 Signal Model of HSMM
  • 8.6 Infinite HSMM and HDP-HSMM
  • 8.7 HSMM Versus HMM
  • 8.7.1 HMM Using HSMM Algorithms
  • 8.7.2 HSMM Using HMM Algorithms
  • 9 Applications of HSMMs
  • 9.1 Speech Synthesis
  • 9.1.1 Speech Synthesis and ML Estimation of Observations
  • 9.1.2 Other Applications Similar to Speech Synthesis
  • 9.2 Human Activity Recognition
  • 9.3 Network Traffic Characterization and Anomaly Detection
  • 9.4 fMRI/EEG/ECG Signal Analysis
  • References
  • Back Cover.

Ejemplares similares