Tighter Expected Generalization Error Bounds via Wasserstein Distance

Neural Information Processing Systems 

This work presents several expected generalization error bounds based on the Wasserstein distance.