In this thesis, we propose a recurrent FIR neural network, develop a constrained formulation for neural network learning, study an e_cient violation guided backpropagation algorithm for solving the constrained formulation ...