Value Determination with General Function Approximators

Vassilis Papavassiliou and Stuart Russell

EECS Department, University of California, Berkeley

Technical Report No. UCB/CSD-98-1005

1998

This publication is archived. It is kept only for reference purposes, so it is no longer being updated and may not meet accessibility standards. If you need this content in a different format, please email webteam@eecs.berkeley.edu.

A new algorithm is described for value determination in Markov decision processes. The algorithm works with arbitrary approximate representations of the value function. We show that if the approximating family is agnostically PAC-learnable, then the algorithm converges to a solution that is close to the globally optimal solution in the approximating family.

BibTeX citation:

@techreport{Papavassiliou:CSD-98-1005,
    Author= {Papavassiliou, Vassilis and Russell, Stuart},
    Title= {Value Determination with General Function Approximators},
    Year= {1998},
    Month= {May},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/1998/6411.html},
    Number= {UCB/CSD-98-1005},
    Abstract= {A new algorithm is described for value determination in Markov decision processes. The algorithm works with arbitrary approximate representations of the value function. We show that if the approximating family is agnostically PAC-learnable, then the algorithm converges to a solution that is close to the globally optimal solution in the approximating family.},
}

EndNote citation:

%0 Report
%A Papavassiliou, Vassilis 
%A Russell, Stuart 
%T Value Determination with General Function Approximators
%I EECS Department, University of California, Berkeley
%D 1998
%@ UCB/CSD-98-1005
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/1998/6411.html
%F Papavassiliou:CSD-98-1005