Can Big Population Always Bring Better Optimization Ability to Evolutionary Computation for Large-Scale Optimization?

Abstract

Artificial intelligence (AI) has fast developed nowadays especially in the deep learning field that most of the deep neural networks pursue the better problem-solving ability by making the network models deeper, larger, and more complex. However, too large and too complex AI algorithms/models require too large computational burden, which is not reality in academic researches nor the right way to real human intelligence. Moreover, in another research branch of AI named evolutionary computation (EC), which is inspired by the biological evolution of nature and swarm intelligence behaviours, will the EC algorithms become more efficient with larger and more complex algorithm model when solving complicated optimization problems like the large-scale optimization problems (LSOPs)? To this concern, this paper investigates whether some existing large-scale optimization EC algorithms can further improve their performance in solving LSOPs by only increasing the population size. We select 12 representative algorithms for investigation, including 4 standard EC algorithms and 8 well-known large-scale optimization algorithms. Then, we adopt the widely-used IEEE Congress on Evolutionary Computation (CEC 2010) LSOPs benchmark test suite to compare the performance of the same algorithms with different population sizes. The experimental results show that simply increasing the population size does not necessarily improve the performance of algorithms in solving LSOPs.