Continuous Optimization

Research in continuous optimization has been active at DIAG since its foundation. Early research was essentially devoted to the theory of exact penalization and to the development of algorithms for the solution of constrained nonlinear programming problems through unconstrained techniques.

Significant early contributions were also given in the field of unconstrained optimization, with the introduction of non monotone line searches, non monotone globalization strategies and convergent derivative-free line search techniques. The Continuous Optimization group later expanded into an active and highly valued optimization research team with a wide range of interests.

The following areas are object of current research.

Exact penalty and augmented Lagrangian methods, still constituting the founding block of many optimization methods and a springboard for many of the studies of the group.

Non-monotone methods and decomposition techniques for the solution of difficult large-scale nonlinear optimization problems and nonlinear equations.

Preconditioning Newton-Krylov and Nonlinear Conjugate Gradient methods in nonconvex large scale optimization, which is an important tool for efficiently solv- ing large difficult problems.

Derivative-free algorithms, of special interest in many engineering applications where even the calculation of function values is problematic and very time-con- suming.

Global optimization, which is an essential tool for solving problems where local non-global solutions may be meaningless.

Semidefinite programming, that plays an essential role in the development of effi- cient algorithms for solving relaxations of non-convex and integer problems.

Finite dimensional variational inequalities and complementarity problems, which often arise in modelling a wide array of real-world problems where competition is involved.

Generalized Nash equilibrium problems, which are emerging as a winning way of looking at several classical and non-classical engineering problems.

Training methods for neural networks and support vector machines, for construct- ing surrogate models of complex systems from sparse data through learning tech- niques.

Mixed Integer Nonlinear Programming (MINLP) problems that combine combina- torial aspects with nonlinearities.

The Continuous Optimization group interacts intensively with many other research groups, both in the academic and industrial world, in an ongoing cross-fertilization pro- cess. This process led to several innovative applications in such different fields as:

Design of electro-mechanic devices.

Development of electromagnetic diagnostic equipments.

Power allocation in TLC.

Shape optimization in ship design.

Multiobjective optimization of nanoelectronic devices.

Optimization of ship itineraries for a cruise fleet.

Sales forecasting in retail stores.

Moreover, as a spin-off of the activity carried out in applied optimization, the company ACTOR (Analytics, Control Technologies and Operations Research) has been founded. ACTOR is participated by Sapienza University, by researchers of the Depart- ment and by the private company ACT Solutions. The main aim of ACTOR is to develop and commercialize advanced optimization models and methods to be employed in the production and management of goods and services.

 

People

Linee di ricerca

Big Data Optimization
Bilevel Optimization
Derivative Free Methods
Engineering Design Optimization
Game Engineering
Global Optimization
Mixed Integer Nonlinear Programming
Neural Networks and Support Vector Machines
Nonlinear Optimization
Parallel and distributed optimization methods
Resource allocation in communication networks
Semidefinite Programming
Simulation–based optimization
Variational Inequalities

© Università degli Studi di Roma "La Sapienza" - Piazzale Aldo Moro 5, 00185 Roma