This paper is intended for a technical audience, and assumes familiarity with Reinforcement Learning. An algorithm described within, called VGL(1), contains the theory used to create the first demo.
PDF download: Reinforcement Learning by Value Gradients. The paper is hosted at Arxiv.org.
Call for help: I would welcome feedback. Particularly, are the arguments sound and convincing? Is the proof in Appendix A sound? Please email me with any comments. I am trying to get this paper up to the standard of being accepted for journal publication. Thanks for any feedback and your efforts in reading it. Michael Fairbank. March 2008.
NEW: I've written some accompanying web-pages to explain the theory more intuitively. This demo does not assume full familiarity with Reinforcement Learning theory, so is intended to be much easier to understand. I'd welcome comments on that too.