Artificial Evolution by Viability Rather than Competition

Abstract

Evolutionary algorithms are widespread heuristic methods inspired by natural evolution to solve difficult problems for which analytical approaches are not suitable. In many domains experimenters are not only interested in discovering optimal solutions, but also in finding the largest number of different solutions satisfying minimal requirements. However, the formulation of an effective performance measure describing these requirements, also known as fitness function, represents a major challenge. The difficulty of combining and weighting multiple problem objectives and constraints of possibly varying nature and scale into a single fitness function often leads to unsatisfactory solutions. Furthermore, selective reproduction of the fittest solutions, which is inspired by competition-based selection in nature, leads to loss of diversity within the evolving population and premature convergence of the algorithm, hindering the discovery of many different solutions.Here we present an alternative abstraction of artificial evolution, which does not require the formulation of a composite fitness function. Inspired from viability theory in dynamical systems, natural evolution and ethology, the proposed method puts emphasis on the elimination of individuals that do not meet a set of changing criteria, which are defined on the problem objectives and constraints.Experimental results show that the proposed method maintains higher diversity in the evolving population and generates more unique solutions when compared to classical competition-based evolutionary algorithms. Our findings suggest that incorporating viability principles into evolutionary algorithms can significantly improve the applicability and effectiveness of evolutionary methods to numerous complex problems of science and engineering, ranging from protein structure prediction to aircraft wing design.

Gaining Insight into Quality Diversity

“…Fitness QD, like fitness itself, may be deceptive and lead to the false conclusion that one algorithm is outperforming another, when in fact it discovers fewer different, successful solutions. (3) Viability Search [7], which like the fitness and random variants is not informed by any BC, can often achieve better performance than other methods which are not informed by the BC that is used to calculate QD. (4) Different BCs drive the search in different ways, and counter-intuitively algorithms driven by different BCs from which QD is calculated on can outperform variants that were actually informed by the BC used to calculate QD (compare how MAP-Elites run with either FullTrajectoryBC or HalfTrajectoryBC rate on Success QD in the top-right plot).…”

“…In the simulations 2 , we model the kinematics of an e-puck On each maze variant, we run five search algorithms: MAP-Elites [9], Novelty Search [6], Viability Evolution [7], random search, and a standard fitness-based Genetic Algorithm (GA). The latter is tested in two different configurations of selection pressure, i.e.…”

“…An alternative algorithm, which has not been previously studied in the context of Quality Diversity is Viability Evolution (ViE) [7]. ViE represents an alternative abstraction of artificial evolution, which, similar to non-objective search, does not require the formulation of an explicit fitness function.…”

Viability Principles for Constrained Optimization Using a (1+1)-CMA-ES

“…Viability Evolution [8,9] is an abstraction of artificial evolution that models an optimization process using viability boundaries, which are modified over time to drive the search towards desirable regions of a search space, as shown in Figure 1. Under this abstraction, mutations can produce viable solutions, which survive, or non-viable solutions, which are eliminated from the population.…”

A multi-cycled sequential memetic computing approach for constrained optimisation

“…In [23], a novel variant of invasive weed optimization was combined as a local refinement procedure within differential evolution [23]. The combination of variability evolution [36] and CMA-ES [37] was proposed in [38] for the NLP.…”