Connectionists: PhD position in Reinforcement Learning/Planning

[Olivier Buffet]

PhD position in Reinforcement Learning/Planning
at INRIA (Nancy, France):

  Title:
Planning under Uncertainty by Direct Policy Search

  Summary:
Automating the control of discrete event systems is an important issue
in a variety of domains: logistics, management of natural ressources,
autonomous embedded systems... This is all the more difficult that the
system's dynamics is uncertain:
- the outcomes of commands depend on external factors;
- the system may be modelled approximately; or
- observations of the system state are partial and noisy.
Such problems can be addressed through the field of probabilistic planning.

Classical probabilistic planning algorithms compute the utility
(=expected distance to the goal) of each action in each accessible
state, which becomes intractable when the state and action spaces grow.
This PhD position follows a different direction. The candidate will work
  on recent and promising algorithms based on optimizing a controller
(such as a neural network). The first planner of this type (the
Factored-Policy Gradient planner (FPG)) won the probabilistic track of
the International Planning Competition 2006.

  Keywords:
probabilistic planning, Reinforcement Learning, Markov Decision
Processes, stochastic optimization, policy-gradient

For more information and for applications, go to:
http://www.talentsplace.com/syndication1/inria/ukdoc/details.html?id=PNGFK026203F3VBQB6G68LOE1&LOV5=4509&LOV2=4490&LOV6=4513&LG=EN&Resultsperpage=20&nPostingID=2418&nPostingTargetID=5800&option=52&sort=DESC&nDepartmentID=28 


About INRIA PhD opportunities:
http://www.inria.fr/travailler/opportunites/doc.en.html