A Variational Learning Algorithm for the Abstract Hidden Markov Model

Abstract

We present a fast algorithm for learning the parameters of the abstract hidden Markov model, a type of hierarchical activ-ity recognition model. Learning using exact inference scales poorly as the number of levels in the hierarchy increases; therefore, an approximation is required for large models. We demonstrate that variational inference is well suited to solve this problem. Not only does this technique scale, but it also offers a natural way to leverage the context specific indepen-dence properties inherent in the model via the fixed point equations. Experiments confirm that the variational approx-imation significantly reduces the time necessary for learning while estimating parameter values that can be used to make reliable predictions.