Chapter 15 Probabilistic Reasoning over Time
Chapter 15, Sections 1-5 Outline Time and uncertainty Inference: ltering, prediction, smoothing Hidden Markov models Kalman lters (a brief mention) Dynamic Bayesian networks Particle ltering
Time and uncertainty The world changes; we need to track and predict it Diabetes management vs vehicle diagnosis Basic idea: copy state and evidence variables for each time step
Markov processes (Markov chains)
Example
Inference tasks
Filtering
Filtering example
Smoothing
Smoothing example
Most likely explanation
Viterbi example
Hidden Markov models
Country dance algorithm
Kalman lters
Updating Gaussian distributions
Simple 1-D example
General Kalman update
2-D tracking example: ltering
2-D tracking example: smoothing
Where it breaks
Dynamic Bayesian networks
DBNs vs. HMMs
DBNs vs Kalman lters
Exact inference in DBNs
Likelihood weighting for DBNs
Particle ltering
Particle ltering contd.
Particle ltering performance
Chapter 15, Sections 1-5 Summary
Chapter 15, Section 6 Outline Speech as probabilistic inference Speech sounds Word pronunciation Word sequences
Speech as probabilistic inference
Phones
Speech sounds
Phone models
Phone model example
Word pronunciation models
Isolated words
Continuous speech
Language model
Combined HMM
DBNs for speech recognition
Chapter 15, Section 6 Summary Since the mid-1970s, speech recognition has been formulated as probabilistic inference Evidence = speech signal, hidden variables = word and phone sequences "Context" effects (coarticulation etc.) are handled by augmenting state Variability in human speech (speed, timbre, etc., etc.) and background noise make continuous speech recognition in real settings an open problem