Using Relative Novelty to Identify Useful Temporal Abstractions in Reinforcement Learning

Authors

Özgür Şimşek, University of Massachusetts - Amherst
Andrew G. Barto, University of Massachusetts - Amherst

Publication Date

2004

Abstract

We present a new method for automatically creating useful temporal abstractions in reinforcement learning. We argue that states that allow the agent to transition to a different region of the state space are useful subgoals, and propose a method for identifying them using the concept of relative novelty. When such a state is identified, a temporallyextended activity (e.g., an option) is generated that takes the agent efficiently to this state. We illustrate the utility of the method in a number of tasks.

Comments

This paper was harvested from CiteSeer

Recommended Citation

Şimşek, Özgür and Barto, Andrew G., "Using Relative Novelty to Identify Useful Temporal Abstractions in Reinforcement Learning" (2004). Computer Science Department Faculty Publication Series. 5.
Retrieved from https://scholarworks.umass.edu/cs_faculty_pubs/5