SWITZERLAND (Lausanne and Zinal)

Selected items

• During my third year at the university, I introduce myself to algorithmic, combinatoric and complexity challenges. My principal dream was (and remains!) to bring answers to some crucial questions yet unsolved, like those related to the theory of NP-Completeness and the design of time optimal parallel algorithms. 

• In my master report, I have developped a simple work optimal algorithm for sorting items in linear time. The mains idear was to allocate each stage of the classical heapsort to one processor and perform pipeline operations. The main advantage that makes this algorithm good for pratical issue (ATM network) was the addressing rule and the possibility to performs several operations in the same cycle. I continued this work in my first PhD year and we were about to validate it when another team discovered nearly the same algorithm and published it.

• Another interesting thing I have done during my master was to built a software (named SYSTOLICA) for the design and simulation of systolic algorithms.

• My first contact with IRISA was based in the previous sorting algorithm together with a challenge to built an efficient parallel algorithm for the Kronecker product and for the Algebraic Path Problem.

• I received the IEEE price of the best communication in the conference renpar11 (see here). Note that it was my first conference participation

• The two years following my phd, I introduced myself to large-scale convex optimization and its relation with combinatorial optimization. Very exciting topic and efficient approach with serious challenges.

• At EMBL, I was involved in computing research activities in synchrotron crystallography. It has been proven that specific radiation damage can help in solving macromolecular crystal structures. However, global damage can also occur at the same time, thereby producing a joint effect which can be hindering depending on its magnitude and the distribution of the B-values. In order to keep handling the benefit of the specific damage, we need to deconvolute between these two effects and thereby provide an efficient intensities approximation of the susceptible atoms subset. Moreover, because data collection is subject to several perturbating factors (conditions and aleas of experiments), refinement techniques need to be developped in order to either correct the data, rescale in order to make them internally consistent, or estimate critical parameters. A conjunction of such factors may lead to serious mathematical and numerical challenges

• I am currently involved in high performance computing research activities (multicore, accelerator-based, and standard parallel (super)computers).