Multitask Optimization Based on Distribution Matching

Evolutionary multitasking optimization (EMTO), a newly developed approach in evolutionary computation (EC), simultaneously solves multiple tasks by transferring effective information through cross-task knowledge transfer (KT). In traditional EMTO algorithms, KT is typically implemented by directly performing crossover between individuals from the source and target populations. However, the approach may not always be appropriate, as different populations have distinct distributions. Therefore, this study develops a novel multitask optimization algorithm based on distribution matching (DMMTO), which incorporates a distribution matching (DM) strategy that matches the distributions of the source and target populations to ensure that transferred individuals from the source are better suited to the target, as well as a simple random crossover (SRC) strategy that enhance knowledge exchange within populations, thereby ensuring effective KT. Experimental outcomes from the CEC2017 multitask benchmark show that the DMMTO algorithm significantly surpass other advanced algorithms, thereby confirming its effectiveness.