Artificial intelligenceLesson 2 of 2 in the course Reasoning under uncertainty

Hidden Markov Models: Structure, Inference, Learning, and Extensions

5 flashcards · 5 quiz questions · 5 match pairs · 5 fill-in sentences · English

Save course
Lesson progress
0 / 5

Your progress is kept in this browser only.Sign up to save it

Summary

A hidden Markov model (HMM) describes a process whose changing states cannot be observed directly, while observable outputs provide clues about those states. This lesson introduces the model's central idea and the assumptions that make it useful.

An HMM contains two related processes. The hidden process, often written X1,X2,…X_1, X_2, \ldots, moves through states over time. The observation process, written Y1,Y2,…Y_1, Y_2, \ldots, produces outputs that can be seen. The states are called hidden states because they are not directly available to the observer.

All flashcards

What does an HMM transition matrix represent?
Probabilities of moving from each hidden state to each possible next state.
What does the forward algorithm compute?
The probability of an observation sequence, efficiently summing over hidden-state paths.
What is the Viterbi algorithm used for?
Finding the single most likely complete hidden-state sequence given observations.
What does Baum–Welch estimate?
HMM parameters through expectation-maximization, typically finding a local maximum-likelihood solution.
What is a factorial HMM?
A model in which observations depend on multiple hidden Markov chains.

Quiz questions

  1. Which algorithm efficiently estimates an earlier hidden state using the complete observation sequence, including later observations?Answer: The forward-backward algorithm
  2. An HMM has 4 hidden states and categorical transitions. How many independent transition parameters are there?Answer: 12
  3. Which statement best describes the Baum–Welch algorithm?Answer: It applies expectation-maximization to find a local maximum-likelihood solution.
  4. If a factorial HMM has K hidden chains with N states each, how many joint state configurations are possible?Answer: NKN^K
  5. Which model is described as avoiding the label-bias problem associated with maximum entropy Markov models?Answer: A linear-chain conditional random field

Based on the Wikipedia article Hidden Markov model, licensed CC BY-SA 4.0. Adapted into a summary, flashcards and quizzes.

More in Artificial intelligence

All Artificial intelligence lessons
Artificial intelligenceA* Search: Pathfinding, Heuristics, Guarantees, and Variants5 flashcards · 5 quiz questions · 4 gamesPlay free →Artificial intelligencePDDL Core Structure, Version History, and Planning Extensions5 flashcards · 5 quiz questions · 4 gamesPlay free →Artificial intelligenceDescription Logic: Knowledge Bases, ALC Semantics, and Reasoning5 flashcards · 5 quiz questions · 4 gamesPlay free →Artificial intelligenceBayesian Networks: Inference, Structure, and Learning5 flashcards · 5 quiz questions · 4 gamesPlay free →Artificial intelligenceOntologies in Information Science: Models, Engineering, and Applications5 flashcards · 5 quiz questions · 4 gamesPlay free →Artificial intelligenceMinimax and Maximin: Decision Theory, Games, and Adversarial Search5 flashcards · 5 quiz questions · 4 gamesPlay free →