Don't Bet on Luck Alone:

Enhancing Behavioral Reproducibility of Quality-Diversity Solutions in Uncertain Domains

GitHub

ArXiv

In this work, we introduce Archive Reproducibility Improvement Algorithm (ARIA); a plug-and-play approach that improves the reproducibility of the solutions present in an archive. We propose it as a separate optimization module, relying on natural evolution strategies, that can be executed on top of any Quality-Diversity (QD) algorithm. Our module mutates solutions to (1) optimize their probability of belonging to their niche, and (2) maximize their fitness. The performance of our method is evaluated on various tasks, including a classical optimization problem and two high-dimensional control tasks in simulated robotic environments. We show that our algorithm enhances the quality and descriptor space coverage of any given archive by at least 50%.

Ant Omni-directional Task

Illusory Archive returned by MAP-Elites

The star represents the selected solution to evaluate several times (those reevaluations are shown on the side)

Evaluation 1

Evaluation 2

Evaluation 3

Same Archive after Optimization by ARIA for Enhanced Reproducibility

Evaluation 1

Evaluation 2

Evaluation 3

Walker2D Uni-directional Task

Example 1 - High-performing Solution

Illusory Archive returned by PGA-ME

Evaluation 1

Evaluation 2

Evaluation 3

Same Archive after Optimization by ARIA for Enhanced Reproducibility

Evaluation 1

Evaluation 2

Evaluation 3

Example 2 - Average Solution

Illusory Archive returned by PGA-ME

Evaluation 1

Evaluation 2

Evaluation 3

Same Archive after Optimization by ARIA for Enhanced Reproducibility

Evaluation 1

Evaluation 2

Evaluation 3

Another archive generated by ARIA - ES Init

Unlike the above results, this variant of ARIA has not been initialized with a QD-generated archive.

Instead, it has been initialized with a single individual resulting from an Evolution Strategy optimization.

Evaluation 1

Evaluation 2

Evaluation 3

Page updated

Google Sites

Report abuse