9 Hidden Markov models

If a rocket blows up, someone’s probably going to get fired, so rocket scientists and engineers must be able to make confident decisions about all components and configurations. They do so by physical simulations and mathematical deduction from first principles. You, too, have solved science problems with pure logical thinking. Consider Boyle’s law: pressure and volume of a gas are inversely related under a fixed temperature. You can make insightful inferences from these simple laws about the world that have been discovered. Recently, machine learning has started to play the role of an important sidekick to deductive reasoning.

Rocket science and machine learning aren’t phrases that usually appear together except unless you literally walk a week in my shoes. (Check out my author bio!) But nowadays, modeling real-world sensor readings by using intelligent data-driven algorithms is more approachable in the aerospace industry. Also, the use of machine-learning techniques is flourishing in the health care and automotive industries. But why?

This influx can be partly attributed to better understanding of interpretable models, which are machine-learning models in which the learned parameters have clear interpretations. If a rocket blows up, for example, an interpretable model might help trace the root cause.

9.1 Example of a not-so-interpretable model

9.2 Markov model

9.3 Hidden Markov model

9.4 Forward algorithm

9.5 Viterbi decoding

9.6 Uses of HMMs

9.6.1 Modeling a video

9.6.2 Modeling DNA

9.6.3 Modeling an image

9.7 Application of HMMs

Summary