Cerfacs Enter the world of high performance ...

Announced
Deadline for registration: 15 days before the starting date of each training
Duration : 3 days / (21 hours)

CANCELLED

 

 

Abstract

In this training course, modern methods for solving optimization problems are detailed. Newton or Quasi-Newton methods for the solution of unconstrained minimization problems are first addressed. Globalization techniques such as trust region methods or adaptive cubic regularization are then detailed. Methods for solving problems without derivatives and problem with general constraints are also outlined. Finally, the solution of nonlinear least-squares problems arising in large-scale inverse problems with application to Earth sciences are reviewed.

Target participants

Engineers, physicists, computer scientists and numerical
analysts who wish to develop basic knowlegde to solve optimisation problems.

Prerequisites

Basic knowledge in linear algebra, numerical analysis and geometry.

Scientific contact : Serge GRATTON

Fee

• Trainees/PhDs/PostDocs : 150 €
• CERFACS shareholders/CNRS/INRIA : 450 €
• Public : 900 €

Program

(Every day from 9h to 17h30)

Day 1

1. Examples of industrial optimization problems.
2. Crucial points for optimization problems modeling: characteristics of the cost function and constraints, importance of the convexity, scaling of the variables and curse of dimensionality for global optimization.
3. Optimality conditions for unconstrained optimization problems.
4. Hands on exercises in Matlab
5. Reverse amphi: the participants introduce their optimization issues and the training team proposes possible relevant solution methods.

Day 2

1. Theory of Lagange multipliers for constrained optmisation.
2. Optimisation methods using interior or exterior penalty approaches and projection approaches.
3. Hands on session in Matlab: augmented Lagrangian method.
4. Reverse amphi: the participants introduce their optimization issues and the training team proposes possible relevant solution methods.

Day 3

1. Derivative free optimisation in 1D.
2. Generalization and introduction to model-based and direct-search methods.
3. Hands on exercises in Matlab.