**An Integrated Neuroevolutionary Approach to Reactive Control and High-level Strategy**. Nate
Kohl and Risto Miikkulainen. *IEEE Transactions on Evolutionary Computation*,
2011. To appear

One promising approach to general-purpose artificial intelligence isneuroevolution, which has worked well on a number of problems fromresource optimization to robot control. However, state-of-the-artneuroevolution algorithms like NEAT have surprising difficulty onproblems that are fractured, i.e. where the desired actions changeabruptly and frequently. Previous work demonstrated that bias andconstraint (e.g. RBF-NEAT and Cascade-NEAT algorithms) can improvelearning significantly on such problems. However, experiments in thispaper show that relatively unrestricted algorithms (e.g. NEAT) stillyield the best performance on problems requiring reactive control.Ideally, a single algorithm would be able to perform well on bothfractured and unfractured problems. This paper introduces such analgorithm called SNAP-NEAT that uses adaptive operator selection tointegrate strengths of NEAT, RBF-NEAT, and Cascade-NEAT. SNAP-NEATis evaluated empirically on a set of problems ranging from reactivecontrol to high-level strategy. The results show that SNAP-NEAT canadapt intelligently to the type of problem that it faces, thus layingthe groundwork for learning algorithms that can be applied to a widevariety of problems.

@Article{kohl:ieeetec11, author = "Nate Kohl and Risto Miikkulainen", title = "An Integrated Neuroevolutionary Approach to Reactive Control and High-level Strategy", journal = "IEEE Transactions on Evolutionary Computation", year = "2011", note = "To appear", abstract = { One promising approach to general-purpose artificial intelligence is neuroevolution, which has worked well on a number of problems from resource optimization to robot control. However, state-of-the-art neuroevolution algorithms like NEAT have surprising difficulty on problems that are fractured, i.e. where the desired actions change abruptly and frequently. Previous work demonstrated that bias and constraint (e.g. RBF-NEAT and Cascade-NEAT algorithms) can improve learning significantly on such problems. However, experiments in this paper show that relatively unrestricted algorithms (e.g. NEAT) still yield the best performance on problems requiring reactive control. Ideally, a single algorithm would be able to perform well on both fractured and unfractured problems. This paper introduces such an algorithm called SNAP-NEAT that uses adaptive operator selection to integrate strengths of NEAT, RBF-NEAT, and Cascade-NEAT. SNAP-NEAT is evaluated empirically on a set of problems ranging from reactive control to high-level strategy. The results show that SNAP-NEAT can adapt intelligently to the type of problem that it faces, thus laying the groundwork for learning algorithms that can be applied to a wide variety of problems. }, bib2html_pubtype = {Journal}, bib2html_rescat = {Machine Learning} }

Generated by bib2html.pl (written by Patrick Riley ) on Sun May 15, 2011 08:27:15