We reformulated the problem of training the neural networks model into a convex optimization problem by performing a local quadratic expansion of the cost function and adding the necessary constraints. We designed a new ...