John Asmuth

About me
  • I am a fourth year student in the PhD program at Rutgers University.
  • I am a member of RL3, Michael Littman's Rutgers Laboratory for Real Life Reinforcement Learning.

    Research Interests
  • Bayesian Reinforcement Learning

    Publications
  • John Asmuth and Lihong Li and Michael L. Littman and Ali Nouri and David Wingate, A Bayesian Sampling Approach to Exploration in Reinforcement Learning, Proceedings of The 25th Conference on Uncertainty in Artificial Intelligence (UAI-09), 2009.
  • John Asmuth and Michael L. Littman and Robert Zinkov, Potential-based Shaping in Model-based Reinforcement Learning, Proceedings of the Twenty-Third AAAI Conference on Artificial Intelligence (AAAI-08), 2008.
  • Hakan L. S. Younes and Michael L. Littman and David Weissman and John Asmuth, The First Probabilistic Track of the International Planning Competition, Journal of Artificial Intelligence Research 24, Pages 851-887, 2005.

    Current Work
  • I am currently working on "ROAR: Relocatable Outcomes Across Regions", a framework for continuous state RL that clusters the space based on locality and dynamics.
  • I find Bayesian approaches to clustering very interesting.

    Past Work
  • I was on the tech committee for RL comp 2009
  • Ninjas Vs Pirates - a game created for my AI class in fall '06. I did the graphics and much of the user interface.
  • In fall '06 I was a TA for cs105: Great Insights in Computer Science and cs503: Computational Thinking. In spring '07 I was a TA for cs314: Principles of Programming Languages. From fall '08 to spring '09 I was a TA for cs113: Software Methodology.

    Useful Documents
    Here is a paper bank on Bayesian RL prepared by Chris Mansley.
    Things you can do with matrices.
    Things you can do with gaussians.
    Things you can do with latex.
    The wikipedia page on conjugate priors.

    Useful Libraries (by others)
    scons is way better than make.
    RL-Glue is a multi-platform multi-language library for interfacing RL agents and environments.
    gzstream.tgz is a C++ library for gzip compression onto a stream.
    mdpsim is a C++ simulator for probabilistic planning.

    Useful Libraries (by me)
    strxml.tgz is a C++ library for parsing xml off of a stream.
    diastream.tgz is a C++ library for writing dia files. Requires gzstream.
    cpputil.tgz is a set of C++ utility classes and functions.
    javawn is a java interface to WordNet

    Contact
    My office is Hill 420, my lab is Hill 427.

    John Asmuth's Facebook profile