Hidden Markov Model in TensorFlow
##Jupyter Notebook: Check out the Notebook for Examples.
- Efficient way of finding the most likely state sequence.
- Method is general statistical framework of compound decision theory.
- Maximizes a posteriori probability recursively.
- Assumed to have a finite-state discrete-time Markov process.
- The goal of the forward-backward algorithm is to find the conditional distribution over hidden states given the data.
- It is used to find the most likely state for any point in time.
- It cannot, however, be used to find the most likely sequence of states (see Viterbi)
- Expectation Maximization Inference of unknown parameters of a Hidden Markov Model.
Unrolled graph
Belief Propagation
Backtrack
Forward-Backward
Re-estimate
Unrolled
