The backward algorithm computes suffix likelihoods in a Hidden Markov model. With , its recurrence is . Together with the forward algorithm, it gives posterior state probabilities proportional to . The terminal condition must include any explicit End-state transition.
New to topics? Read the docs here!