A set of MATLAB & MEX/C functions one can use to build basic static & dynamic probabilistic models. Current PMT provides support for the following probabilistic models: Gaussian mixtures, Factor analyzers, Markov chains, Hidden Markov models, and Linear dynamic systems. For each probabilistic model, PMT provides functions for simulation (sampling from the model), inference (hidden state estimation), and learning of model parameters from data. PMT supports multiple inference methods, both exact and approximate (e.g., winner takes all), based on the Bayesian network equivalence of the model. Model parameters are learned from data using maximum likelihood estimation (EM). PMT also supports arbitrary distributions of training data, something that comes useful in building recursive additive mixtures of those models (e.g., boosting).
For any questions about PMT please contact vladimir+pmt AT cs.rutgers.edu.