Go up to Question 4
Solution to Question 3
A* is used for path planning problems where the aim is to find the
shortest path to a goal (the length of a path is the sum of the
lengths of the arcs) and where there is heuristic information in terms
of an estimate of the distance from a node to a goal node.
In neural network learning, the aim is to assign values to parameters
that minimize the error. To use A* this problem has to be converted
to a path planning problem.
The way presented in the text to convert a constraint satisfaction
problem to a search problem, namely where nodes correspond to
assignments of values to (some of the) variables, and the neighbors
correspond to an assignment of values to another variable, doesn't
work because
- There is no value until all of the variables have been assigned
values. (I.e., there are no arc costs except for the last arc).
- There is no heuristic function.
- Even if there was, the search space is too big for systematic search.