We consider the problem of finding the minimal ε-equivalent MDP for an MDP given in its tabular form. We show that the problem is NP-Hard and then give a bicriteria approximation algorithm to the problem. We suggest that the right measure for finding minimal ε-equivalent model is L1 rather than L∞ by giving both an example, which demonstrates the drawback of using L∞, and performance guarantees for using L1. In addition, we give a polynomial algorithm that decides whether two MDPs are equivalent.
|Number of pages||14|
|Journal||Lecture Notes in Computer Science|
|State||Published - 2003|
|Event||16th Annual Conference on Learning Theory and 7th Kernel Workshop, COLT/Kernel 2003 - Washington, DC, United States|
Duration: 24 Aug 2003 → 27 Aug 2003