Abstract: Hidden Markov models (HMMs) are an important tool for data exploration and engineering applications. A very large number of real-world problems can be framed and "solved" using the HMM framework (speech recognition, inferring the "true" price of an instrument over time, text segmentation, etc.)

I will review HMMs and then sweep through a large number of interesting generalizations that lead up to a state-of-the-art unifying model my collaborators and I call the infinite structured explicit duration hidden Markov model (ISEDHMM). Joint work with Mike Dewar (Bitly) and Jonathan Huggins (Columbia)