OBJECTIVE FUNCTION

With both architectures we want to minimize

$$E^p = \sum_p \sum_{k=1}^{n+1} \frac{1}{2} (eval(s^p(k-1),s^p(k)))^2.$$ (3)

In words, we wish to find a sequence of subgoals such that the sum of the costs of all involved subprograms is minimized. This will be done by using gradient descent techniques to be described in the next section.

Back to Subgoal learning - Hierarchical Learning
Pages with Subgoal learning pictures