global convergence

Global convergence means that an optimization algorithm is guaranteed to find a global minimum of the loss function, regardless of the initialization, which is a desirable characteristic for training deep learning models.

13 papers