Discriminative State Space Models

In this work, we introduce and study Discriminative State-Space Models (DSSMs) . We provide the precise mathematical definition of this class of models in Section 2 . Roughly speaking, a DSSM follows the same general structure as in ( 1) and consists of a state predictor g and an observation predictor h . However, no assumption is made about the form of the stochastic process used to generate observations. This family of models includes existing generative models and other state-based discriminative models (e.g. RNNs) as special cases, but also consists of some novel algorithmic solutions explored in this paper.