Learning predictive representations from a history

Predictive State Representations (PSRs) have shown a great deal of promise as an alternative to Markov models. However, learning a PSR from a single stream of data generated from an environment remains a challenge. In this work, we present a formalism of PSRs and the domains they model. This formalization suggests an algorithm for learning PSRs that will (almost surely) converge to a globally optimal model given sufficient training data.

• Type: Conference
• Year: 2005
• Where: ICML
• Authors: Eric Wiewiora
• Keywords: ICML 2005, Machine Learning, Predictive State Representations, Sufficient Training Data