Illuminating search spaces by mapping elites
Date: 10th July 2025
arXiv link
Key Point
- Aims to find the best solution from each neighbourhood of the space.
- Tells us about how the best performance changes with different dimensions.
- Method:
- Uses evolutionary algorithm. Feature space is description of system in meaningful dimensions.
- User selects dimensions of interest in which we are interested to see how performance varies.
- Creates grid in this space
- Creates random solutions… identifies which grid space they belong in
- Mutates them and sees if the new mutant outperforms the version in its grid space.
- Result is a set of best solutions at each point in the space