Genetic algorithms (GAs) are widely used in multi-objective optimization for solving complex problems. There are two distinct approaches for GA design: generational and steadystate algorithms. Most of the current state-of-the-art GAs are generational, although there is an increasing interest to steady-state algorithms as well. However, for algorithms based on non-dominated sorting, most of steady-state implementations have higher computation complexity than their generational counterparts, which limits their applicability.We present a fast implementation of a steady-state version of the NSGA-II algorithm for two dimensions. This implementation is based on a data structure which has O(N ) complexity for single solution insertion and deletion in the worst case. The experimental results show that our implementation works noticeably faster than steady-state NSGA-II implementations which use fast non-dominated sorting.

No abstract

No abstract

In parallel and distributed environments, generational evolutionary algorithms often do not exploit the full potential of the computation system since they have to wait until the entire population is evaluated before starting selection procedures. Steady-state algorithms are often seen as a solution to this problem, since fitness evaluation can be done by multiple threads in an asynchronous way. However, if the algorithm updates its state in a complicated way, the threads will eventually have to wait until this update finishes. State update procedures that are computationally expensive are common in multiobjective evolutionary algorithms.We have implemented an asynchronous steady-state version of the NSGA-II algorithm. Its most expensive part, non-dominated sorting, determines the time needed to update the state. We turned the existing incremental non-dominated sorting algorithm into an asynchronous one using several concurrency techniques: a single entry-level lock, finer-grained locks working with non-domination levels, and a non-blocking approach using compare-and-set operations. Our experimental results reveal the trade-off between the work-efficiency of the algorithm and the achieved amount of parallelism.