Table 16.

Advantages and disadvantages of neural networks for optimization.

Algorithm	Advantages	Disadvantages
Artificial Neural Networks (ANNs)	excel in optimizing problems involving nonlinear relationships among variables performance and problem optimization are optimized in an adaptable way through learning able to generalize data never before seen by them and use this for optimization convenient for handling huge amounts of data allow for massive parallelization of computing indispensable in accurately approximating functions with any arbitrary continuous relationship ∙	suffer from lack of transparency (black box nature) difficult tuning of hyperparameters (number of layers and nodes, learning rate, etc.) resource-intensive need huge amounts of data for accurate training computationally demanding prone to overfitting (may fail to generalize well for new datasets and search spaces)
Convolutional Neural Networks (CNNs)	well suited for image-related optimization tasks and sequential data with spatial hierarchies able to recognize patterns regardless of their position in an image (translation invariance) automatic recognition and extraction of features during optimization offers transfer learning (using training in one search space for another) lower risk of overfitting thanks to parameter sharing across spatial dimensions	need huge amounts of data for accurate training computationally demanding resource-intensive suffer from lack of transparency (black box nature) •minuscule perturbations of the input data can result in incorrect predictions vulnerable to adversarial attacks, which negatively reflects on security and robustness