All my publications are [here]

See also [HAL]

My books

A selection of recent contributions dealing with environmental issues

Trajectory and research topics

Classical scheduling and Packing

I was working on scheduling for many years. The main contributions concerned the extensions of scheduling models and algorithms in order to take into account communications and the design of optimization algorithms on interconnection networks. I was interested in designing low-cost approximation algorithms for scheduling parallel jobs (moldable and malleable jobs) and packing problerms.

Implementing parallel applications

I was interested in implementing efficiently large parallel applications on the successive generations of parallel supports. Originally, such applications were mainly reduced to regular numerical algorithms, then, following the evolution of parallel platforms, the target applications highly diversified. The size of the problems also increase and the race of always better performances will never end at a price always heavier for the environment.

Adaptive algorithms

The apparition of new computing platforms highly changed the way of designing algorithms. One of the main feature that is requested today is that the algorithms can adapt them-selves to versatile conditions. It is possible to adapt existing algorithms for selecting adequate algorithms depending on the considered data, but a more challenging problem is to determine automatically how to combine good algorithms. Another view of this problem is to study and design algorithms that can absorb the effects of disturbances or incertainties in an evolvingi setting. Several approaches are possible from sensitivity analysis to robustness. I believe that flexibility is the right way: Starting from an initial solution computed with estimated data, some correction mechanisms are added at run-time for being able to react to disturbances.

Multi-objective optimization

There exist many efficient techniques for solving classical combinatorial problems. Today, we are looking for solutions of problems with diversified objectives. I studied classical multi-objective problems for determining one "good" trade-off solution (approximation of the Zenith). We are looking now for approximation algorithms of the whole Pareto set or at least an approximation of it. I proposed to study how to adapt the existing methods for many objectives, including unexpected dimensions linked to environmental impacts.

Analysis and optimization of Fault-Tolerance algorithms

While designing parallel algorithms on large scale parallel platforms, we can not assume that they will run without failures. I studied such problems in the perspective of bi-objective optimization (for determining the best trade-off between performance and reliability). We proposed a new probabilistic model for computing the expected execution time with checkpoint-restart mechanisms. The most efficient solution today is rather to slow-down and develop relilient algorithms that do not require a too high level of reliability.