
Gradient networks (GradNets) are novel neural network architectures that directly parameterize and learn gradients of various function classes, including gradients of convex functions (mGradNets). These networks exhibit specialized architectural constraints to ensure correspondence to gradient functions, enabling efficient parameterization and robust theoretical guarantees.


coremsg

gradient-networks-parameterizing-and-learning-gradients-of-functions


Gradient Networks: Parameterizing and Learning Gradients of Functions
