New Pathways in Coevolutionary Computation

TL;DR

The paper surveys coevolutionary computation and presents two novel algorithms, OMNIREP and SAFE, to advance how representations and objectives are handled in evolving systems. OMNIREP jointly optimizes a representation and its encoding via two coevolving populations, using cross-evaluation to yield effective solutions across multiple problem classes. SAFE introduces a commensalistic coevolution framework that evolves both solutions and objective functions, using objective-function novelty to guard against conflating the objective with its measure, and demonstrates benefits in robotic maze navigation and multiobjective optimization. Together, the work provides proof-of-concept that representation/encoding discovery and objective-function discovery can be automated and beneficial, with code available for reuse.

Abstract

The simultaneous evolution of two or more species with coupled fitness -- coevolution -- has been put to good use in the field of evolutionary computation. Herein, we present two new forms of coevolutionary algorithms, which we have recently designed and applied with success. OMNIREP is a cooperative coevolutionary algorithm that discovers both a representation and an encoding for solving a particular problem of interest. SAFE is a commensalistic coevolutionary algorithm that maintains two coevolving populations: a population of candidate solutions and a population of candidate objective functions needed to measure solution quality during evolution.

Figures (5)

• Figure 1: Coevolution: (a) cooperative: Purple-throated carib feeding from and pollinating a flower (credit: Charles J Sharp, https://commons.wikimedia.org/wiki/File:Purple-throated_carib_hummingbird_feeding.jpg); (b) competitive: predator and prey---a leopard killing a bushbuck (credit: NJR ZA, https://commons.wikimedia.org/wiki/File:Leopard_kill_-_KNP_-_001.jpg); (c) commensalistic: Phoretic mites attach themselves to a fly for transport (credit: Alvesgaspar, https://en.wikipedia.org/wiki/File:Fly_June_2008-2.jpg).
• Figure 2: Fitness computation in OMNIREP, where two populations coevolve, one comprising representations, the other encodings. Fitness is computed by combining a representation individual (R) with an encoding individual (E).
• Figure 3: In a maze problem a robot begins at the start square and must make its way to the goal square (objective). Shown above are paths (green) of robots evolved by a standard evolutionary algorithm with fitness measured as distance-to-goal, evidencing how conflating the objective with the objective function leads to a non-optimal solution.
• Figure 4: A single generation of SAFE vs. a single generation of a standard evolutionary algorithm. The numbered circles identify sequential steps in the respective algorithms. The objective function can comprise a single or multiple objectives.
• Figure 5: Solutions to the maze problems, evolved by SAFE.