Constrained deep learning is an advanced approach to training deep neural networks by incorporating domain-specific constraints into the learning process.

Learn dynamical systems as a difference of convex functions (DC) using a feedforward Neural Network (NN) architecture with DC structure. The resulting model learns the dynamics f in DC form as follows: f= f1 -f2 where f1 and f2 are convex functions. The DC structure of the network allows to independently express f1 and f2 as two input convex NN.