Crowding

Crowding is a replacement strategy in evolutionary computation that inserts a new offspring into the population only if it is sufficiently different from the individuals it most resembles, thereby preventing premature convergence on a single solution. Unlike fitness sharing, which operates on the selection step by penalizing similarity, crowding operates at the replacement step by maintaining diversity through local competition. The standard implementation — deterministic crowding — pairs each offspring with the most similar parent and replaces the parent only if the offspring is fitter.

The method was introduced by De Jong (1975) and has been extended in numerous variants (restricted tournament selection, probabilistic crowding, multi-niche crowding). Its biological analogue is the process by which newly arrived individuals in an ecosystem are excluded from established territories unless they can outperform the incumbent. Crowding treats the population not as a single breeding pool but as a collection of local neighborhoods, each with its own competitive dynamic. The technique is essential for preserving multiple solutions in multimodal optimization and has been applied to multi-objective optimization where the goal is not a single optimum but a diverse Pareto front.