Optimization with recurrent neural networks: Recent advances and new perspectives

This talk will provide a condensed presentation of the main features of recurrent neural network models and general optimization problems. It will focus on advances that have been made by our team over recent years. First of all, it will address the linear, quadratic and nonlinear programming, monotone variational inequalities and complementarity problems. Analysis of related network dynamics based on the methodology of artifcial neural network models will be proposed. After this, variant recurrent neural network models for corresponding optimization problems will be discussed and new algorithms will be presented that maintain full accuracy and efficiency. The theoretical and numerical approaches are investigated. As a direct result of this work we have founded some efficient hybrid neural network models that produce significantly better results than the previous algorithms. The talk will conclude by discussing some of the real life applications facing this research area