Standard genetic algorithms (GAs) are known to work well in certain cases, and to suffocate in others. For example, when a task has several component parts, the results may be disappointing, as they may tend to evaluate the best on one part of the task, while ignoring the others.

In this paper, an extension of this paradigm, the group GA, is presented. While the authors struggle to provide a biological parallel to the approach and the algorithms presented, the results show that group GAs work well. The example used in this paper is an immune system matching task.

The niching or speciation may or may not be known a priori. Both cases are discussed in depth. The authors state that they “can [now] evolve groups of individuals to collectively perform tasks with minimal a priori knowledge of how many subtasks there are or how they should be shared.”

This paper is an extension of a conference paper. It is concise, but well elaborated where necessary, and is supported by background knowledge and actual experiments. This gives the reader confidence that this GA extension work needs be tested in cases where the task at hand may be too complex for standard GAs to handle due to the number of subtasks.