Sorry for the late reply; I was busy with some personal work.

Yeah it is true, We do not need to target global minima as it is not that easy to reach there.

Additionally, other techniques such as regularization, dropout, and data augmentation can also help prevent overfitting and promote better generalization.

In short, while finding the global minimum is a theoretical goal in optimization, it's not always necessary for developing neural networks that generalize well