CMU-ML-12-108
Machine Learning Department
School of Computer Science, Carnegie Mellon University



CMU-ML-12-108

Spectral Approaches to Learning
Predictive Representations

Byron Boots

September 2012

Ph.D. Thesis

CMU-ML-12-108.pdf


Keywords: System Identification, Reinforcement Learning, Spectral Learning, Predictive State Representations, Kernel Methods


A central problem in artificial intelligence is to choose actions to maximize reward in a partially observable, uncertain environment. To do so, we must obtain an accurate environment model, and then plan to maximize reward. However, for complex domains, specifying a model by hand can be a time consuming process. This motivates an alternative approach: learning a model directly from observations. Unfortunately, learning algorithms often recover a model that is too inaccurate to support planning or too large and complex for planning to succeed; or, they require excessive prior domain knowledge or fail to provide guarantees such as statistical consistency. To address this gap, we propose spectral subspace identification algorithms which provably learn compact, accurate, predictive models of partially observable dynamical systems directly from sequences of action-observation pairs. Our research agenda includes several variations of this general approach: spectral methods for classical models like Kalman filters and hidden Markov models, batch algorithms and online algorithms, and kernel-based algorithms for learning models in high- and infinite-dimensional feature spaces. All of these approaches share a common framework: the model's belief space is represented as predictions of observable quantities and spectral algorithms are applied to learn the model parameters. Unlike the popular EM algorithm, spectral learning algorithms are statistically consistent, computationally efficient, and easy to implement using established matrixalgebra techniques. We evaluate our learning algorithms on a series of prediction and planning tasks involving simulated data and real robotic systems.

175 pages


SCS Technical Report Collection
School of Computer Science