Estimating Hidden Markov Models (HMMs)

Estimating Hidden Markov Models (HMMs)

Recap of Hidden Markov Models (HMMs)

We don’t see state

Each state produces a visible output. this output is drawn from a distribution for each state.

We observe a sequence of outputs, not states.

Estimating HMMs with the Viterbi algorithm

Assume we know transition matrix. and starting probls

Given we observe sequence of outputs, what were most likely actual paths?

Virbiti returns this

Estimating HMMs with the forward algorithm

Given we have observed outputs, what is the chance of being in a certain state at a certain time?

Estimating HMMs with the forward-backward algorithm

We calculate state x at time t given all obs.

Baum-Welch algorithm

Kalman filters