@inproceedings{37f6087fade24f41996061a72fa2a5df,
title = "Axioms for rational reinforcement learning",
abstract = "We provide a formal, simple and intuitive theory of rational decision making including sequential decisions that affect the environment. The theory has a geometric flavor, which makes the arguments easy to visualize and understand. Our theory is for complete decision makers, which means that they have a complete set of preferences. Our main result shows that a complete rational decision maker implicitly has a probabilistic model of the environment. We have a countable version of this result that brings light on the issue of countable vs finite additivity by showing how it depends on the geometry of the space which we have preferences over. This is achieved through fruitfully connecting rationality with the Hahn-Banach Theorem. The theory presented here can be viewed as a formalization and extension of the betting odds approach to probability of Ramsey and De Finetti [Ram31, deF37].",
keywords = "Banach Space, Linear Functional, Probability, Rationality, Utility",
author = "Peter Sunehag and Marcus Hutter",
year = "2011",
doi = "10.1007/978-3-642-24412-4_27",
language = "English",
isbn = "9783642244117",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
pages = "338--352",
booktitle = "Algorithmic Learning Theory - 22nd International Conference, ALT 2011, Proceedings",
note = "22nd International Conference on Algorithmic Learning Theory, ALT 2011 ; Conference date: 05-10-2011 Through 07-10-2011",
}