Evolutionary Multitasking Optimization Enhanced by Geodesic Flow Kernel

In an era of parallel computing, evolutionary multitasking optimization (EMT) has become a popular optimization paradigm due to its ability to optimize several tasks simultaneously. The common knowledge can improve the solving quality and efficiency for each component optimization task when transferred among tasks. Therefore, the performances of traditional EMT algorithms mostly rely on the correlation between tasks. In the field of EMT, a key issue needing to be solved urgently is the impact of negative transfer when tackling optimization tasks with low correlation. In order to overcome the short board of this situation, this paper proposes a multiobjective EMT algorithm EMT-GFK. In the proposed algorithm, a union subspace of the optimization tasks is designed to extract the compact information. Furthermore, the geodesic flow kernel based domain adaptation is applied to learn a nonlinear mapping matrix, which can increase the correlation between tasks. The numerical experiments and results analysis on the MO-MTO test suits demonstrate the effectiveness of proposed EMT-GFK.