...[Cormen et al. 1990].
Although [Cormen et al. 1990] presents DAG-SP only for deterministic acyclic problems, it applies straightforwardly to the stochastic case.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...backgammon
Though backgammon is not strictly acyclic, loops are quite rare, so ROUT may work well anyway.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...ROUT.
Unlike TD, which works only with parametric function approximators for which tex2html_wrap_inline2038 can be calculated, ROUT can work with arbitrary function approximators, including batch methods such as projection-pursuit and locally weighted regression. For these comparative experiments, however, we used linear or neural net fits for both algorithms.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

Justin A. Boyan
Sat Jun 22 20:49:48 EDT 1996