next up previous
Next: Bibliography Up: subgoalsicann Previous: An Experiment: Learning to

Conclusion

By giving a constructive example it has been demonstrated that so-called `higher-level symbolic processes' (in our case sub-goal generation) and so-called `sub-symbolic' neural networks can be made compatible. This may be viewed as one first step towards neural network architectures that bridge the gap between `sub-symbolic' and `symbolic' computation. A quite different approach to adaptive sub-goaling has been described recently [11].



Juergen Schmidhuber 2003-03-14

Back to Subgoal learning - Hierarchical Learning
German pages with Subgoal learning pictures