Department of mathematical engineering

Université catholique de Louvain

I am an FNRS research fellow (aspirant FNRS) at the Unversité catholique de Louvain, in the department of mathematical engineering (part of the ICTEAM) until Sept. 2014. There, I obtained my Ph.D. on February 14, 2014 with the advisory of Pierre-Antoine Absil and Vincent Blondel.

I look into applications of **optimization methods on
manifolds** (for which I develop a
toolbox called Manopt) to solve computational problems.
A reference in this field is the book Optimization
Algorithms on Matrix Manifolds. My initial research in
that area was about the design and analysis of numerical
algorithms for curve fitting
on manifolds, which I started during my master
thesis. More recently, I have been working on **low-rank
matrix completion** and **synchronization of
rotations**, which are at the center of my doctoral
thesis. Over the last three years, I visited PACM
(the program in applied and computational mathematics at
Princeton University) a number of times, to collaborate with Amit Singer
on synchronization problems.

I hold a master's degree in mathematical engineering from the
Ecole
polytechnique de Louvain (*ingénieur civil en
mathématiques appliquées*) (2008-2010). During my
master's, I completed a one month internship in video
processing under the supervision of Benoît
Macq at UCLouvain's TELE lab. (2008) and a three months
internship in mathematical seismology with Laurent
Demanet in the mathematics departments at Stanford then
MIT (2009).

Here is a CV.

Manopt, available at manopt.org,
is a user-friendly, open source and **documented **Matlab
toolbox which can be used to leverage the power of modern
Riemannian optimization algorithms with ease. Manopt won the ORBEL
Wolsey Award 2014 for best open source operational
research implementation.

Synchronization is the problem of estimating elements $g_1, \ldots, g_N$ in a group $G$, given measurements of relative quantities: $h_{ij} \approx g_i^{}g_j^{-1}$. These elements are best visualized on a graph (undirected), where each element $g_i$ is a node and there exists an edge between two nodes $g_i$ and $g_j$ if a measurement $h_{ij}$ is available. I focus on $G = \mathrm{SO}(n)$, the group of rotations:

$$\mathrm{SO}(n) = \{ R \in \mathbb{R}^{n\times n} \colon R^TR = I_n \ \mathrm{ and } \ \operatorname{det}(R) = +1 \}.$$

SynchronizeMLE is the distribution of Matlab codes for this project, available under BSD license. It contains code both to perform the estimation and to compute Cramér-Rao bounds.

Let $M \in \mathbb{R}^{m\times n}$ be a matrix with low rank $r \ll \min(m, n)$. Low-rank matrix completion is the task of estimating (or recovering) $M$ from measurements $\hat M_{ij} \approx M_{ij}$ of a few entries $(i, j) \in \Omega$.

At NIPS 2011, we proposed RTRMC, a Riemannian trust-region method for low-rank matrix completion:

In this project, which was the topic of my master's
thesis, we are given time-labeled points on a Riemannian
manifold $\mathcal{M}$ (for example, on a sphere, on the group
of rotations, on the set of positive-definite matrices, etc.):
$p_1, \ldots, p_n$, associated to timestamps $t_1 \leq \ldots
\leq t_n$. The goal is to propose a curve (a model) on the
manifold, $\gamma \colon [t_1, t_n] \to \mathcal{M}$, **such
that the curve fits the data** (exactly for
interpolation, reasonably for regression): $\gamma(t_i)
\approx p_i$ **and **such that $\gamma$ **is
smooth** in some suitable sense. Interpolation and
regression are fundamental operations in signals processing.
They serve the goals of** denoising and resampling
acquired data**. These tasks are well understood when
the data belongs to a Euclidean space such as $\mathbb{R}^n$,
but much less so **when the data belon****g****s
to a nonlinear manifold**.

Av. George Lemaitre 4-6, bte L4.05.01

1348 Louvain-la-Neuve, Belgium

Phone: +32 10 47 80 10

Office: Euler a-124 (level -1)

E-mail: nicolas.boumal@uclouvain.be

My officemate is Romain
Hollanders.

Romain, Karim and I won 2nd place in the SIAM Math Matters: Apply It! 2011 contest, with our poster introducing the large public to the importance of math in monitoring fetal ECG's.

My Erdös number is 3, courtesy of my co-author and advisor Vincent Blondel.

Research will get you places! It got me in: Palo Alto, Boston, Princeton, London, Prague, Cannes, Lisbon, Milan, Dagstuhl, Granada, Sierra Nevada, Valencia, Berlin, Les Houches, Costa da Caparica, Paris, Florence... and various places in Belgium (Louvain-la-Neuve, Leuven, Liège, La Roche, Mons, Knokke, Daverdisse...).

Romain, Karim and I won 2nd place in the SIAM Math Matters: Apply It! 2011 contest, with our poster introducing the large public to the importance of math in monitoring fetal ECG's.

My Erdös number is 3, courtesy of my co-author and advisor Vincent Blondel.

Research will get you places! It got me in: Palo Alto, Boston, Princeton, London, Prague, Cannes, Lisbon, Milan, Dagstuhl, Granada, Sierra Nevada, Valencia, Berlin, Les Houches, Costa da Caparica, Paris, Florence... and various places in Belgium (Louvain-la-Neuve, Leuven, Liège, La Roche, Mons, Knokke, Daverdisse...).

- Mathématiques 1 (FSAB1101), TA, autumn 2008, autumn 2009
- Projet 1 (FSAB1501), TA, autumn 2010
- Théorie des Matrices (INMA2380), TA, spring 2011, autumn 2013
- Signaux et Systèmes (LFSAB1106), TA, autumn 2011
- Analyse numérique : approximation, interpolation, intégration (LINMA2171), TA, autumn 2011 and 2012
- Mathématiques 2 (LFSAB1102), TA, spring 2012
- Modélisation et analyse des systèmes dynamiques (LINMA2370), TA, autumn 2012
- Projet en ingénierie mathématique (LINMA2360), TA, spring 2012 and 2013
- Projet en mathématiques appliquées (LINMA1375), TA, spring 2013
- Systèmes dynamiques non linéaires (LINMA2361), TA, autumn 2013