Next: An Experiment: Learning to
Up: Cascaded sub-goaling with a
Previous: Cascaded sub-goaling with a
Perhaps the simplest answer to this question is: With a given task,
first make a try without any sub-goal. If this does not work, try
one sub-goal, then two, etc.
An extension of the trial-and-error approach
would be to train a fourth network to map start/goal
combinations to numbers of sub-goals. The training signals could
be obtained by the trial-and-error-method: For a given problem,
the desired output of the fourth network should be a representation of the
minimal number of necessary sub-goals.
Back to Subgoal learning - Hierarchical Learning
German pages with Subgoal learning pictures