Residual Learning Inspired Crossover Operator and Strategy Enhancements for Evolutionary Multitasking