In contrast, HOLO's nonlinear activation functions can adaptively adjust the optimization path during training, allowing the algorithm to converge more efficiently to the global optimum.