Neural Dynamics Gradient Time Acceleration Optimizer

NDGTA optimizer guides parameters of neural networks toward flat minima during the training for generalization.

GENERALIZATION denotes the capability of deep neural networks (DNNs) to execute tasks on unseen data and embodies a form of intelligence. Improving model generalization has attracted extensive attention in recent years.

NDGTA optimizer embeds the loss function within a neural dynamics model to promote convergence to a flat minimum.

This implementation refer to the paper NDGTA.