What is it about?
We have introduced the concept of behavior landscapes, in an analogy to fitness landscapes. Furthermore, We have explored the influence of different diversity metrics and levels of evolutionary pressure on evolvability in the context of neuroevolutionary divergent search. The results indicate that even relatively small amounts of evolutionary pressure promote evolvability and that the Gaussian kernel density-based metric offers competitive performance.
Featured Image
Photo by John Schaidler on Unsplash
Why is it important?
Evolvability is highly important as evolvable solutions may be only a few mutations/steps of gradient ascents away from a set of diverse solutions, each of which performs well under different environmental conditions. This is particularly important in problems involving dynamic fitness landscapes (i.e. environments with non-stationary conditions). Furthermore, evolvable solutions are more likely to spread around the space of possible phenotypes, hence expediting the evolutionary process. Demonstrations that divergent search promotes evolvability corroborate the importance of this type of search in replicating important properties of biological evolution and in solving a wide array of practical problems that require many diverse solutions.
Read the Original
This page is a summary of: On Evolvability and Behavior Landscapes in Neuroevolutionary Divergent Search, July 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3583131.3590427.
You can read the full text:
Resources
Contributors
The following have contributed to this page