Lab Home | Phone | Search | ||||||||
|
||||||||
We discuss approximate solution of large systems of equations, based on projection on a low dimensional subspace and simulation. The key advantage of this approach is that all required linear algebra operations are low-dimensional (the dimension of the approximation subspace). Furthermore, the methods are well-suited for parallel computation. Hence, extremely large systems can be addressed. Our methods are motivated by recent advances in approximate dynamic programming, and extend the class of temporal difference methods, used for policy evaluation within this context. We will discuss this connection, and overview the associated convegence properties. Host: Frank Alexander |