Clement Maria

Clément Maria

INRIA Sophia Antipolis - Méditerranée :

2004 route des Lucioles -- BP 93
06902 Sophia Antipolis Cedex
FRANCE
Office: Y303, Byron building
Phone: +33 4 92 38 78 93

E-Mail :

Clement [dot] Maria [at] inria [dot] fr

I am a PhD student within the GEOMETRICA group from INRIA, under the supervision of Jean-Daniel Boissonnat.

My current researches focus on the design of algorithmic technics for higher dimensional geometry.

Articles and Preprints:

An Exponantial Lower Bound on the Complexity of Regularization Paths (with Bernd Gaertner and Martin Jaggi ).

Education:

2010-2011	Master Parisien de la Recherche en Informatique, Paris, FRANCE (link).
	École Normale Supérieure de Cachan, FRANCE (link).
	Second year of Master in Computer Science
2009-2010	Erasmus exchange student in the University of Aarhus, DENMARK (link).
	École Normale Supérieure de Cachan, FRANCE.
	First year of Master in Computer Science
2008-2009	École Normale Supérieure de Cachan, FRANCE.
	Bachelor of Computer Science, with Honors: mention Bien
	Bachelor of Mathematics, with Honors: mention Assez Bien
2008	Successfull candidate at the prestigious competitive examination of École Normale Supérieure de Cachan.
2006-2008	Classes Préparatoires aux Grandes Écoles Lycée Masséna, Nice, FRANCE.
	Mathematics, Physics and Computer Science (MPSI/MP*)
2006	Baccalauréat Scientifique, with Honors: Mention Très Bien

Research Internships:

July-August 2010

Internship in the Emo Welzl Group, Theory of Combinatorial Algorithms, of Institute of Theoretical Computer Science, ETH Zuerich, SWITZERLAND.

Supervized by Bernd Gaertner

Subject: An Exponential Lower Bound on the Complexity of Regularization Paths.

Abstract: For a variety of regularization methods, algorithms computing the entire solution path have been developed recently. Solution path algorithms do not only compute the solution for one particular value of the regularization parameter but the entire path of solutions, making the selection of an optimal parameter much easier. It has been assumed that these piecewise linear solution paths have only linear complexity, i.e. linearly many bends.
Gaertner, Giesen, and Jaggi have announced a construction for an exponentially long (w.r.t. number of bends) solution path of a support vector machine. However, the construction is informal. The goal of this project is to adapt the construction and proove mathematically that it would lead to a solution path with exponantially many bends.

June-July 2009

Internship in the GEOMETRICA team of INRIA Sophia Antipolis-Mediterranee, FRANCE.

Supervized by Jean-Daniel Boissonnat

Subject: A Compact data Structure to Represent the Delaunay Triangulation

Abstract: The GEOMETRICA team is working on geometry algorithms for dimension higher than 3. This technics concern many fields: space-time (d=4), phase space (d=6), machine learning (any dimension). The main issue is due to the fact that the size of the data structures depends exponantially in the dimension. The team has developped an incremental algorithm for the Delaunay triangulation construction in arbitrary dimension.
The goal of this project is to replace the adjacency list representation of the Delaunay graph with a more efficient representation. This would permit to store bigger triangulations and/or work on higher dimensions.
The idea is to adapt the algorithm of Blandford and al. for graph compression to our problem. We will study both static and dynamic cases.

Last Update : September 2011