1: \begin{abstract}

2: The Hidden Markov Model (HMM) is one of the mainstays of statistical

3: modeling of discrete time series, with applications including speech

4: recognition, computational biology, computer vision and econometrics.

5: Estimating an HMM from its observation process is often addressed via

6: the Baum-Welch algorithm, which is known to be susceptible to local

7: optima.  In this paper, we first give a general characterization of

8: the basin of attraction associated with any global optimum of the

9: population likelihood.  By exploiting this characterization, we

10: provide non-asymptotic finite sample guarantees on the Baum-Welch

11: updates, guaranteeing geometric convergence to a small ball of radius

12: on the order of the minimax rate around a global optimum.  As a

13: concrete example, we prove a linear rate of convergence for a hidden

14: Markov mixture of two isotropic Gaussians given a suitable mean

15: separation and an initialization within a ball of large radius around

16: (one of) the true parameters.  To our knowledge, these are the first

17: rigorous local convergence guarantees to global optima for the Baum-Welch algorithm in

18: a setting where the likelihood function is nonconvex.  We complement

19: our theoretical results with thorough numerical simulations studying

20: the convergence of the Baum-Welch algorithm and illustrating the

21: accuracy of our predictions.

22: \end{abstract}

23: