id: 05980301
dt: a
an: 05980301
au: Kemmerich, Thomas; Kleine Büning, Hans
ti: On the power of global reward signals in reinforcement learning.
so: Klügl, Franziska (ed.) et al., Multiagent system technologies. 9th German
    conference, MATES 2011, Berlin, Germany, October 6‒7, 2011.
    Proceedings. Berlin: Springer (ISBN 978-3-642-24602-9/pbk). Lecture
    Notes in Computer Science 6973. Lecture Notes in Artificial
    Intelligence, 53-64 (2011).
py: 2011
pu: Berlin: Springer
la: EN
cc: 
ut: reinforcement learning; global reward; conceptual models; partial
    observability
ci: 
li: doi:10.1007/978-3-642-24603-6_7
ab: Summary: Reinforcement learning is investigated in various models,
    involving single and multiagent settings as well as fully or partially
    observable domains. Although such models differ in several aspects,
    their basic approach is identical: agents obtain a state observation
    and a global reward signal from an environment and execute actions
    which in turn influence the environment state. In this work, we discuss
    the role of such global reward signals. We present a concept that does
    not provide a visible environment state but only offers a numerical
    engineered reward. It will be proven that this approach has the same
    computational complexity and expressive power as ordinary fully
    observable models, but allows to infringe assumptions in models with
    partial observability. To avoid such infringements, we then argue that
    rewards, besides a true reward value, shall never contain additional
    polynomial time decodable information.
rv: