SLIDE 1
CS 730/730W/830: Intro AI
■ Break HMMs
Break
■ Break HMMs
CS 730/730W/830: Intro AI Break HMMs 1 handout: slides final blog - - PowerPoint PPT Presentation
CS 730/730W/830: Intro AI Break HMMs 1 handout: slides final blog - - PowerPoint PPT Presentation
CS 730/730W/830: Intro AI Break HMMs 1 handout: slides final blog entries were due Wheeler Ruml (UNH) Lecture 27, CS 730 1 / 8 Break Wed May 2: HMMs, unsupervised learning, applications Break Mon May 7: special guest Scott
SLIDE 2
SLIDE 3
Hidden Markov Models
■ Break HMMs ■ Models ■ The Model ■ Viterbi Decoding ■ Random ■ EOLQs
Wheeler Ruml (UNH) Lecture 27, CS 730 – 3 / 8
SLIDE 4
Probabilistic Models
■ Break HMMs ■ Models ■ The Model ■ Viterbi Decoding ■ Random ■ EOLQs
Wheeler Ruml (UNH) Lecture 27, CS 730 – 4 / 8
MDPs: Naive Bayes: k-Means: Markov chain: Hidden Markov model:
SLIDE 5
A Hidden Markov Model
■ Break HMMs ■ Models ■ The Model ■ Viterbi Decoding ■ Random ■ EOLQs
Wheeler Ruml (UNH) Lecture 27, CS 730 – 5 / 8
P(xt = j) =
- i
P(xt−1 = i)P(xt = j|xt−1 = i) P(et = k) =
- i
P(xt = i)P(e = k|x = i) More concisely: P(xt) =
- xt−1
P(xt−1)P(xt|xt−1) P(et) =
- xt
P(xt)P(e|x)
SLIDE 6
Viterbi Decoding
■ Break HMMs ■ Models ■ The Model ■ Viterbi Decoding ■ Random ■ EOLQs
Wheeler Ruml (UNH) Lecture 27, CS 730 – 6 / 8
given: transition model T(s, s′) sensing model S(s, o)
- bservations o1, . . . , oT
find: most probable s1, . . . , sT initialize S × T matrix v with 0s v0,0 ← 1 for each time t = 0 to T − 1 for each state s for each new state s′ score ← vs,t · T(s, s′) · S(s′, ot) if score > vs′,t+1 vs′,t+1 ← score best-parent(s′)← s trace back from s with max vs,T
SLIDE 7
Random
■ Break HMMs ■ Models ■ The Model ■ Viterbi Decoding ■ Random ■ EOLQs
Wheeler Ruml (UNH) Lecture 27, CS 730 – 7 / 8
applications unsupervised learning: dimensionality reduction
SLIDE 8
EOLQs
■ Break HMMs ■ Models ■ The Model ■ Viterbi Decoding ■ Random ■ EOLQs