From: Parag (parag_at_cs.washington.edu)
Date: Fri May 09 2003 - 11:02:40 PDT
Incremental Pruning: A Simple, Fast, Exact Method for Partially Observable
Markov Decision Processes
-- A. Cassandra, M. Littman, N. Zhang
The paper improves upon the Incremental Pruning algorithm
for solving POMDPs, and performs a set of experiments to
compare the performance of Incremental Pruning with other
algorithms.
The main contribution of the paper is to give an algorithm
which, as the authors claim, improves upon the basic incremental
pruning method. The algorithm does this by exploiting some kind of
regularity in the problem while performing the purge.
The authors also perform a set of experiments to show
that IP and in particular their improved version perform
better than some of the other known algorithms.
I had a couple of problems while reading the paper.
First, the notation in the beginning seemed a bit
confusing. Second, it was not very clear what was
the objective of the paper - to compare the IP (which
is a known algorithm) with other known algorithms or to
propose an extension to IP which would perform better.
This also casts some doubts on what exactly is the contribution
of the paper is.
As the authors themselves point out, it is not exactly clear
why/where their modification works better than IP. Though,
authors propose this as a future work, I felt that the paper
would have seemed much more convincing had they given some
insight into this.
This archive was generated by hypermail 2.1.6 : Fri May 09 2003 - 11:02:41 PDT