Baum–Welch algorithm
E880219
The Baum–Welch algorithm is an expectation-maximization method used to train the parameters of hidden Markov models from observed data.
Statements (46)
| Predicate | Object |
|---|---|
| instanceOf |
algorithm
ⓘ
expectation–maximization algorithm ⓘ statistical estimation method ⓘ training algorithm ⓘ |
| appliesTo | hidden Markov model NERFINISHED ⓘ |
| assumes |
Markov property for hidden states
ⓘ
conditional independence of observations given states ⓘ |
| basedOn | expectation–maximization (EM) algorithm NERFINISHED ⓘ |
| category |
Expectation–maximization algorithms
ⓘ
Hidden Markov models NERFINISHED ⓘ Machine learning algorithms ⓘ |
| computes | expected sufficient statistics of hidden state sequences ⓘ |
| convergesTo | local maximum of likelihood ⓘ |
| E-stepUses | forward–backward algorithm NERFINISHED ⓘ |
| field |
machine learning
ⓘ
signal processing ⓘ speech recognition ⓘ statistics ⓘ |
| hasStep |
E-step
ⓘ
M-step ⓘ |
| input | observed sequences ⓘ |
| isSpecialCaseOf | general EM algorithm NERFINISHED ⓘ |
| learningType | unsupervised learning ⓘ |
| M-stepUpdates |
emission probabilities
ⓘ
initial state probabilities ⓘ transition probabilities ⓘ |
| namedAfter |
Leonard E. Baum
NERFINISHED
ⓘ
Lloyd R. Welch NERFINISHED ⓘ |
| optimizationCriterion | likelihood of observed data ⓘ |
| optimizationType | maximum likelihood ⓘ |
| output | estimated HMM parameters ⓘ |
| publicationDecade | 1970s ⓘ |
| relatedTo |
Viterbi algorithm
NERFINISHED
ⓘ
forward–backward algorithm NERFINISHED ⓘ |
| requires | initial parameter guess ⓘ |
| usedFor |
learning emission probabilities in HMMs
ⓘ
learning initial state distribution in HMMs ⓘ learning transition probabilities in HMMs ⓘ maximum likelihood estimation of HMM parameters ⓘ parameter estimation in hidden Markov models ⓘ training hidden Markov models ⓘ unsupervised sequence learning ⓘ |
| usedIn |
biosequence analysis
ⓘ
natural language processing ⓘ speech recognition systems ⓘ time series modeling ⓘ |
Referenced by (1)
Full triples — surface form annotated when it differs from this entity's canonical label.