Interactive PBIL with multiple probability vectors for multimodal optimization with implicit performance indices

Interactive population-based incremental learning (IPBIL) is an effective method to solve optimization problems with implicit performance indices. It can significantly reduce user fatigue compared with interactive evolutionary computation (IEC). However, each run of IPBIL can only find one solution or some similar solutions. Thus it is not suitable for multimodal optimization problems with implicit performance indices. To solve this problem, we propose an IPBIL with multiple probability vectors (IPBIL-MPV) in this work. The key idea is to utilize multiple probability vectors to catch different search directions and thus find more than one solutions. We perform a subjective experiment in which IPBIL-MPV is applied to a fashion design problem. The experimental results show that IPBIL-MPV can find several distinct solutions in a run. Thus it is an effective method to solve multimodal optimization problems with implicit performance indices.