Learning the Linear Dynamical System with ASOS

We develop a new algorithm, based on EM, for learning the Linear Dynamical System model. Called the method of Approximated Second-Order Statistics (ASOS) our approach achieves dramatically superior computational performance over standard EM through its use of approximations, which we justify with both intuitive explanations and rigorous convergence results. In particular, after an inexpensive precomputation phase, the iterations of ASOS can be performed in time independent of the length of the training dataset.

• Type: Conference
• Year: 2010
• Where: ICML
• Authors: James Martens
• Keywords: ICML 2010, Inexpensive Precomputation Phase, Linear Dynamical System, Machine Learning, Superior Computational Performance