François Galea's Homepage

photo

I am a researcher at CEA LIST, in the Architecture, IC Design & Embedded Software Division, in Palaiseau, France.

Here is my résumé (PDF format, in french).

Contact:

francois.galea <at> cea.fr

Research fields:

My research fields are mostly operational research and parallelism.

My current job at CEA list consists in applying operational research techniques to solve problems in software compilation for manycore architectures, where the problems are expressed in languages which naturally enable a high level of parallelism. Problems involved include

Previous works:

My thesis work was concerning the optimization of treatment plans in high dose rate brachytherapy. I was working in collaboration with physicists from the Pitié-Salpétrière hospital in Paris, in the radiotherapy service. While most existing approaches were involving integer programming, I developed a linear programming model to solve applications with good quality results. A have also shown that the existing old “safe” (ie, with scientifically proven safety warranties) catheter implantation schemes were not optimal with regard to modern treatment technology.

My former job as a research engineer at the PRiSM laboratory in Versailles, France, consisted in developing and testing a parallel exact solver for “large” instances of the Three-index Quadratic Assigment Problem (Q3AP). Instance sizes varying from 13 to 16 can be considered as large. This work also required development of parts of the Bob++ framework, for which I have written a MPI-based parallelization, as well as a parallel Mixed Integer Programming solver, featuring generic cut generation and strong branching. This work is part of the ANR CHOC project.

I also worked at INRIA in Saclay, France, where I developed a Parametric Integer Programming (PIP) solver for the Parma Polyhedra Library. This library is heavily used in the GCC Graphite component, which makes use of polyhedral modeling techniques to detect static independencies in source code. Such independencies possibly enable dedicated compiler optimizations like automatic parallelization of loops.

I also had great interest in developing experiments on the Cell Broadband Engine Architecture, a very specific class of hybrid processors which provide great processing power at cheap price, at the cost of designing very specific code for them.

Projects:

©2008-2013 François Galea - Last Modified 06/17/15 16:22 Valid XHTML 1.0! Valid CSS!