Multitasking Evolutionary Algorithm with SVM-Based Knowledge Discriminator

Evolutionary multitasking (EMT) algorithms simultaneously address multiple optimization problems while enhancing problem-solving through knowledge migration. However, inappropriate knowledge transfer can lead to negative consequences, affecting search dynamics and algorithm convergence. In this regard, we propose an SVM-based knowledge discriminator (SVMKD) integrated into the EMT algorithm to reduce the likelihood of negative knowledge transfer. SVMKD utilizes task population experiences to train the SVM classifier, discerning solution quality during knowledge transfer and reducing the probability of negative transfer. The effectiveness of SVMKD is evaluated through a comprehensive empirical study on nine single-objective multitasking benchmark problems.