Scientists from the Siberian Federal University (SibFU), in cooperation with foreign colleagues, have presented an improved version of the well-known linear-variational-inequality primal-dual neural network (LVI-PDNN), which is used to optimize mathematical models in various fields.