Niching by multiobjectivization with neighbor information: Trade-offs and benefits

In this paper we investigate the ability of selection methods to enforce niching on multi modal problems. Using theoretical properties where possible, and relying on a sound experimental analysis, we show that the conventional single-objective optimization and novelty search are extreme cases of selection, striving only for quality or diversity. However, in between these well known cases, there are many more possibilities, of which we review eight (including the aforementioned two). Multiobjective selection approaches provide a well-balanced trade-off' between exploration and exploitation. For the multiobjectivization, we recommend to use nearest-better-neighbor information instead of the common nearest-neighbor approaches.