Overestimation learning with guarantees

Adrien Gauffriau, François Malgouyres, Mélanie Ducoffe

We describe a complete method that learns a neural network which is guaranteed to overestimate a reference function on a given domain. The neural network can then be used as a surr…