This work investigates evolvability of continuous-time recurrent neural networks to support the behavior of model-agents subject to fitness criteria that changes over the evolutionary timescale. A population of agents is alternatingly evolved to perform two tasks with inverted fitness awards. Evidence of evolvability is reported; it is shown that the population locates a region of "meta-fitness" in the landscape in which sub-regions of optimality for each task are easily accessible from one another.

This content is only available as a PDF.
© 2016 Massachusetts Institute of Technology Published under a Creative Commons Attribution 4.0 International (CC BY 4.0) license
2016
MIT Press
This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.