Markov Decision Processes with a New Optimality Criterion.
Abstract
A Markov decision process can be characterized by specifying the following three elements: a Markov process on which a return function and decision structure is placed, an objective function or optimality criterion, and a class of allowable policies or controls. For a given Markov decision process with these three elements suitably defined, the standard problems to investigate are the following: The existence of a policy, within the class of allowable policies, which attains the maximal value of the objective function; The fact that the optimal policy has a simple form; The construction of a finite algorithm to compute the optimal policy. The report discusses these problems for standard Markov decision processes with a new optimality criterion. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1971
- Accession Number
- AD0724753
Entities
People
- Stratton C. Jaquette
Organizations
- Stanford University