From: Bhushan Mandhani (bhushan_at_cs.washington.edu)
Date: Mon Nov 17 2003 - 12:58:40 PST
Summary: The paper describes solving MDP's using algebraic decision
diagrams (ADD's).
Main Ideas:
1. Using ADD's is more efficient than the use of corresponding decision
trees. This claim is backed up by experimental data.
2. ADD's have been used in other domains and are well understood. There
are open avenues for their further optimization, due to which they are a
promising approach to solving MDP's.
Flaws:
1. Mutivalued variables are handled by expressing them as the combination
of several boolean variables. It is not clear how much does this affect
the performance of ADD's.
2. The notation used, and the overall presentation of the algorithm was
not good, making the paper difficult to comprehend at times.
Future Work:
1. Exploring dynamic variable reordering to optimize the algorithm.
2. A more thorough evaluation of the utility of ADD's for solving
real-world problems.
This archive was generated by hypermail 2.1.6 : Mon Nov 17 2003 - 12:58:41 PST