Sequence models: HMM decoding

We discuss the decoding problem, why it's more difficult than computing joint probabilities, but why it still has "optimal substructure" that allows us to use a dynamic programming algorithm. We detail the Viterbi dynamic programming algorithm and characterize its time and space requirements. ── Course & channel links ── Course playlist: Sequence Modeling    • Sequence modeling   ── About the author ── Ben Langmead is a Professor of Computer Science at Johns Hopkins University, where his research spans bioinformatics, computational biology, and data-intensive science. He is the author of Bowtie and Bowtie 2; his group has also developed software like Kraken 2 and resources like recount3 and Index Zone, as well as methods for pangenome indexing and querying, based on e.g. the r-index and move structure. His group's methods have been cited over 130,000 times, and he is the winner of awards including an NSF CAREER award, a Sloan Research Fellowship, the Benjamin Franklin award for contributions to open access, and multiple awards for teaching and mentorship. Ben is the founder and principal of InOrder Labs LLC (https://inorderlabs.com), an expert consulting firm in bioinformatics and computational biology. Channel:    / @benlangmead   Teaching materials: https://langmead-lab.org/teaching.html