Objectives of Programme's Learning Outcomes

• Mobilise a structured set of scientific knowledge and skills and specialised techniques in order to carry out computer and management engineering missions, using their expertise and adaptability.
• Master and appropriately mobilise knowledge, models, methods and techniques specific to computer management engineering.
• Communicate and exchange information in a structured way - orally, graphically and in writing, in French and in one or more other languages - scientifically, culturally, technically and interpersonally, by adapting to the intended purpose and the relevant public.
• Argue to and persuade customers, teachers and boards, both orally and in writing.
• Contribute by researching the innovative solution of a problem in engineering sciences.
• Construct a theoretical or conceptual reference framework, formulate innovative solutions from the analysis of scientific literature, particularly in new or emerging disciplines.
• Develop and implement conceptual analysis, numerical modelling, software implementations, experimental studies and behavioural analysis.
• Communicate, in writing and orally, on the approach and its results in highlighting both the scientific criteria of the research conducted and the theoretical and technical innovation potential, as well as possible non-technical issues.

Learning Outcomes of UE

AA - Multi-Objective Optimization This class serves as an introduction the most fundamentals in Multi-Objective Optimization and introduces representative algorithms (illustrating their working principles and discussing their application scope and performance). After the course, the students must be able to understand why Multi-Objective Optimization methods are needed, understand optimality concept in Multi-Objective Optimization, understand different approaches to solve Multi-Objective Optimization problems and understand basics of choosing and implementing Multi-Objective Optimization methods. AA - First-Order Methods for Large Scale Machine Learning This class serves as an introduction to first-order methods to tackle large-scale ooptimization problems that arise for example in data analysis and machine learning. A particular attention will be given to stochastic gradient methods that are used to compute the weights in deep neural networks.  The students must be able to undertsand the theoretical concepts behind these methods (in particular, convergence issues), and to implement and use them on real-world problems.  AA - Topics in Convex Optimization The main subject of this course is to cover some techniques of convex optimization, a field that has undergone considerable development over the last three decades. This course is based on I-MARO-035 (Linear Optimization) and I-MARO-036 (Nonlinear Optimization). The heart of this course is conic optimization and more specifically semidefinite optimization that allows the modelisation of many problems in applied science.  The students must be able to undertsand the theoretical concepts behind these methods and to implement and use them on real-world problems. 

Content of UE

AA - Multi-Objective Optimization Organization of the class: Introduction and background on Multi-Objective Optimization (search space, objective space, Pareto optimality, Pareto front,...). Several basic and advanced optimization methods are presented to solve Multi-Objective Optimization problems. Some tools are presented to evaluate the performance of Multi-Objective algorithms.  AA - First-Order Methods for Large Scale Machine Learning Organization of the class:  - Introduction and motivation to the use of firt-order methods.  - Optimal first-order methods for convex optimization.  - Stochastic gradient methods.  
- Project: Comparison of first-order methods for solving a classification problem. AA - Topics in Convex Optimization Organization of the class:  - Introduction to conic optimization.  - Second-order cone programming.  - Semidefinite programming - Sum-of-squares.  - Project: modeling and solving problems. 

Prior Experience

Good knowledge of optimization techniques and modelling.

Type of Assessment for UE in Q1

• Presentation and/or works
• Written examination

Q1 UE Assessment Comments

For each AA, the project is the evaluation (gruops of 2 students). An additional written exam may be organised. Attending the class is part of the evaluation. For the UE, it concerns a global evaluation that is calculated as follows : If X = evaluation/20 of AA-Multi-Objective Optimization, Y = evaluation20 of AA-First-Order Methods for Large Scale Machine Learning, Z = evaluation/20 of AA-Topics in Convex Optimization. If Min(X,Y,Z) < 8, then the Global evaluation of  UE = Min(X,Y,Z). If Min(X,Y,Z) >= 8, then the Global evaluation of UE = 0.2 * X + 0.4 * Y + 0.4 * Z

Type of Assessment for UE in Q3

• Presentation and/or works
• Written examination

Q3 UE Assessment Comments

For each AA, the project is the evaluation (gruops of 2 students). An additional written exam may be organised. Attending the class is part of the evaluation. For the UE, it concerns a global evaluation that is calculated as follows : If X = evaluation/20 of AA-Multi-Objective Optimization, Y = evaluation20 of AA-First-Order Methods for Large Scale Machine Learning, Z = evaluation/20 of AA-Topics in Convex Optimization. If Min(X,Y,Z) < 8, then the Global evaluation of  UE = Min(X,Y,Z). If Min(X,Y,Z) >= 8, then the Global evaluation of UE = 0.2 * X + 0.4 * Y + 0.4 * Z

Type of Resit Assessment for UE in Q1 (BAB1)

• N/A

Q1 UE Resit Assessment Comments (BAB1)

Not applicable