Approximate Predictive Representations of Partially Observable Systems

We provide a novel view of learning an approximate model of a partially observable environment from data and present a simple implemenf the idea. The learned model abstracts away unnecessary details of the agent's experience and focuses only on making certain predictions of interest. We illustrate our approach in small computational examples, demonstrating the data efficiency of the algorithm.

• Type: Journal
• Year: 2010
• Where: ICML
• Authors: Monica Dinculescu, Doina Precup
• Keywords: Approximate Model, ICML 2010, Machine Learning, Observable Environment, Small Computational Examples