Nicolas Brosse
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Nicolas Brosse has a total of 17 co-authors. Here are the top co-authors by number of shared publications:
- Eric Moulines
École Polytechnique
7 shared publications - Alain Durmus
CMLA -École normale supérieure Paris-Saclay CNRS Université Paris-Saclay 94235 Cachan France
5 shared publications Alain Durmus
2 shared publications- Sean Meyn
Department of Electrical and Computer Engineering, University of Florida, Gainesville, FL, USA
2 shared publications - Sotirios Sabanis
The University of Edinburgh United Kingdom
2 shared publications Éric Moulines
2 shared publications
Latest Publications
- 3 April 2026
- Published by Agence Bibliographique de l'Enseignement Supérieur
Autour de l'algorithme du Langevin : extensions et applications
Cette thèse porte sur le problème de l'échantillonnage en grande dimension et est basée sur l'algorithme de Langevin non ajusté (ULA).Dans une première partie, nous proposons deux extensions d'ULA et fournissons des garanties de convergence précises pour ces algorithmes. ULA n'est pas applicable lorsque la distribution cible est à support compact; grâce à une régularisation de Moreau Yosida, il est néanmoins possible d'échantillonner à partir d'une distribution suffisamment proche de la distribution cible. ULA diverge lorsque les queues de la distribution cible sont trop fines; en renormalisant correctement le gradient, cette difficulté peut être surmontée.Dans une deuxième partie, nous donnons deux applications d'ULA. Nous fournissons un algorithme pour estimer les constantes de normalisation de densités log concaves à partir d'une suite de distributions dont la variance augmente graduellement. En comparant ULA avec la diffusion de Langevin, nous développons une nouvelle méthode de variables de contrôle basée sur la variance asymptotique de la diffusion de Langevin.Dans une troisième partie, nous analysons Stochastic Gradient Langevin Dynamics (SGLD), qui diffère de ULA seulement dans l'estimation stochastique du gradient. Nous montrons que SGLD, appliqué avec des paramètres habituels, peut être très éloigné de la distribution cible. Cependant, avec une technique appropriée de réduction de variance, son coût calcul peut être bien inférieur à celui d'ULA pour une précision similaire.
- 3 April 2026
- Published by Agence Bibliographique de l'Enseignement Supérieur
Autour de l'algorithme du Langevin : extensions et applications
Cette thèse porte sur le problème de l'échantillonnage en grande dimension et est basée sur l'algorithme de Langevin non ajusté (ULA).Dans une première partie, nous proposons deux extensions d'ULA et fournissons des garanties de convergence précises pour ces algorithmes. ULA n'est pas applicable lorsque la distribution cible est à support compact; grâce à une régularisation de Moreau Yosida, il est néanmoins possible d'échantillonner à partir d'une distribution suffisamment proche de la distribution cible. ULA diverge lorsque les queues de la distribution cible sont trop fines; en renormalisant correctement le gradient, cette difficulté peut être surmontée.Dans une deuxième partie, nous donnons deux applications d'ULA. Nous fournissons un algorithme pour estimer les constantes de normalisation de densités log concaves à partir d'une suite de distributions dont la variance augmente graduellement. En comparant ULA avec la diffusion de Langevin, nous développons une nouvelle méthode de variables de contrôle basée sur la variance asymptotique de la diffusion de Langevin.Dans une troisième partie, nous analysons Stochastic Gradient Langevin Dynamics (SGLD), qui diffère de ULA seulement dans l'estimation stochastique du gradient. Nous montrons que SGLD, appliqué avec des paramètres habituels, peut être très éloigné de la distribution cible. Cependant, avec une technique appropriée de réduction de variance, son coût calcul peut être bien inférieur à celui d'ULA pour une précision similaire.
- Open Access
- 9 June 2023
- Published by Springer Nature in Research Square
Development of novel non-woven Triumfetta cordifolia bast fibers and polylactide fibers biocomposites
This study focuses on the development of needle-punched nonwoven composites reinforced with a Triumfetta cordifolia (TC) bast fiber based on a polylactide (PLA) matrix. Flax and hemp fibers, already known for their use in automotive applications, are used as a reference for comparison purposes. TC bast fibers from the equatorial region of Cameroon were extracted by a water retting process. Nonwoven felts were obtained by mixing TC fibers with PLA (50:50 weight ratio) through carding-napping process. The composite materials were obtained by thermocompression of the nonwoven felts with two thickness ranges (2 mm and 3 mm). The results obtained in tensile and flexural tests showed that composite materials with an optimal thickness of 2 mm displayed better mechanical performances. The anisotropy of the mechanical properties in traction and flexion of these composites was highlighted. It was found that the tensile and flexural mechanical properties in the transverse direction (CD) are always superior to those in the machine direction (MD). Compared to flax or hemp nonwoven-reinforced biocomposites, TC fibers-based nonwoven biocomposites showed slightly lower maximum tensile and flexural stiffness values of 2882.7 MPa and 3908.6 MPa respectively. Thermogravimetric analysis revealed that the addition of the plant fibers increased the thermal stability of PLA and the ash content to 8%.
- 9 June 2023
- Published by Springer Nature in Research Square
Open AccessDevelopment of novel non-woven Triumfetta cordifolia bast fibers and polylactide fibers biocomposites
This study focuses on the development of needle-punched nonwoven composites reinforced with a Triumfetta cordifolia (TC) bast fiber based on a polylactide (PLA) matrix. Flax and hemp fibers, already known for their use in automotive applications, are used as a reference for comparison purposes. TC bast fibers from the equatorial region of Cameroon were extracted by a water retting process. Nonwoven felts were obtained by mixing TC fibers with PLA (50:50 weight ratio) through carding-napping process. The composite materials were obtained by thermocompression of the nonwoven felts with two thickness ranges (2 mm and 3 mm). The results obtained in tensile and flexural tests showed that composite materials with an optimal thickness of 2 mm displayed better mechanical performances. The anisotropy of the mechanical properties in traction and flexion of these composites was highlighted. It was found that the tensile and flexural mechanical properties in the transverse direction (CD) are always superior to those in the machine direction (MD). Compared to flax or hemp nonwoven-reinforced biocomposites, TC fibers-based nonwoven biocomposites showed slightly lower maximum tensile and flexural stiffness values of 2882.7 MPa and 3908.6 MPa respectively. Thermogravimetric analysis revealed that the addition of the plant fibers increased the thermal stability of PLA and the ash content to 8%.
- Open Access
On Last-Layer Algorithms for Classification: Decoupling Representation from Uncertainty Estimation
Uncertainty quantification for deep learning is a challenging open problem. Bayesian statistics offer a mathematically grounded framework to reason about uncertainties; however, approximate posteriors for modern neural networks still require prohibitive computational costs. We propose a family of algorithms which split the classification task into two stages: representation learning and uncertainty estimation. We compare four specific instances, where uncertainty estimation is performed via either an ensemble of Stochastic Gradient Descent or Stochastic Gradient Langevin Dynamics snapshots, an ensemble of bootstrapped logistic regressions, or via a number of Monte Carlo Dropout passes. We evaluate their performance in terms of \emph{selective} classification (risk-coverage), and their ability to detect out-of-distribution samples. Our experiments suggest there is limited value in adding multiple uncertainty layers to deep classifiers, and we observe that these simple methods strongly outperform a vanilla point-estimate SGD in some complex benchmarks like ImageNet.
Open AccessOn Last-Layer Algorithms for Classification: Decoupling Representation from Uncertainty Estimation
Uncertainty quantification for deep learning is a challenging open problem. Bayesian statistics offer a mathematically grounded framework to reason about uncertainties; however, approximate posteriors for modern neural networks still require prohibitive computational costs. We propose a family of algorithms which split the classification task into two stages: representation learning and uncertainty estimation. We compare four specific instances, where uncertainty estimation is performed via either an ensemble of Stochastic Gradient Descent or Stochastic Gradient Langevin Dynamics snapshots, an ensemble of bootstrapped logistic regressions, or via a number of Monte Carlo Dropout passes. We evaluate their performance in terms of \emph{selective} classification (risk-coverage), and their ability to detect out-of-distribution samples. Our experiments suggest there is limited value in adding multiple uncertainty layers to deep classifiers, and we observe that these simple methods strongly outperform a vanilla point-estimate SGD in some complex benchmarks like ImageNet.
- 1 October 2019
- Published by Elsevier in Stochastic Processes and their Applications
- Vol. 129 (10) , 3638-3663
The tamed unadjusted Langevin algorithm
- 1 October 2019
- Published by Elsevier in Stochastic Processes and their Applications
- Vol. 129 (10) , 3638-3663
The tamed unadjusted Langevin algorithm
- Open Access
Diffusion approximations and control variates for MCMC
A new methodology is presented for the construction of control variates to reduce the variance of additive functionals of Markov Chain Monte Carlo (MCMC) samplers. Our control variates are definedthrough the minimization of the asymptotic variance of the Langevin diffusion over a family of functions, which can be seen as a quadratic risk minimization procedure. The use of these control variates is theoretically justified. We show that the asymptotic variances of some well-known MCMC algorithms, including the Random Walk Metropolis and the (Metropolis) Unadjusted/Adjusted Langevin Algorithm, are close to the asymptotic variance of the Langevin diffusion. Several examples of Bayesian inference problems demonstrate that the corresponding reduction in the variance is significant.
Open AccessDiffusion approximations and control variates for MCMC
A new methodology is presented for the construction of control variates to reduce the variance of additive functionals of Markov Chain Monte Carlo (MCMC) samplers. Our control variates are definedthrough the minimization of the asymptotic variance of the Langevin diffusion over a family of functions, which can be seen as a quadratic risk minimization procedure. The use of these control variates is theoretically justified. We show that the asymptotic variances of some well-known MCMC algorithms, including the Random Walk Metropolis and the (Metropolis) Unadjusted/Adjusted Langevin Algorithm, are close to the asymptotic variance of the Langevin diffusion. Several examples of Bayesian inference problems demonstrate that the corresponding reduction in the variance is significant.
- Open Access
The Tamed Unadjusted Langevin Algorithm
In this article, we consider the problem of sampling from a probability measure $\pi$ having a density on $\mathbb{R}^d$ known up to a normalizing constant, $x\mapsto \mathrm{e}^{-U(x)} / \int_{\mathbb{R}^d} \mathrm{e}^{-U(y)} \mathrm{d} y$. The Euler discretization of the Langevin stochastic differential equation (SDE) is known to be unstable in a precise sense, when the potential $U$ is superlinear, i.e. $\liminf_{\Vert x \Vert\to+\infty} \Vert \nabla U(x) \Vert / \Vert x \Vert = +\infty$. Based on previous works on the taming of superlinear drift coefficients for SDEs, we introduce the Tamed Unadjusted Langevin Algorithm (TULA) and obtain non-asymptotic bounds in $V$-total variation norm and Wasserstein distance of order $2$ between the iterates of TULA and $\pi$, as well as weak error bounds. Numerical experiments are presented which support our findings.Open AccessThe Tamed Unadjusted Langevin Algorithm
In this article, we consider the problem of sampling from a probability measure $\pi$ having a density on $\mathbb{R}^d$ known up to a normalizing constant, $x\mapsto \mathrm{e}^{-U(x)} / \int_{\mathbb{R}^d} \mathrm{e}^{-U(y)} \mathrm{d} y$. The Euler discretization of the Langevin stochastic differential equation (SDE) is known to be unstable in a precise sense, when the potential $U$ is superlinear, i.e. $\liminf_{\Vert x \Vert\to+\infty} \Vert \nabla U(x) \Vert / \Vert x \Vert = +\infty$. Based on previous works on the taming of superlinear drift coefficients for SDEs, we introduce the Tamed Unadjusted Langevin Algorithm (TULA) and obtain non-asymptotic bounds in $V$-total variation norm and Wasserstein distance of order $2$ between the iterates of TULA and $\pi$, as well as weak error bounds. Numerical experiments are presented which support our findings. - Open Access
Diffusion approximations and control variates for MCMC
A new methodology is presented for the construction of control variates to reduce the variance of additive functionals of Markov Chain Monte Carlo (MCMC) samplers. Our control variates are defined as linear combinations of functions whose coefficients are obtained by minimizing a proxy for the asymptotic variance. The construction is theoretically justified by two new results. We first show that the asymptotic variances of some well-known MCMC algorithms, including the Random Walk Metropolis and the (Metropolis) Unadjusted/Adjusted Langevin Algorithm, are close to the asymptotic variance of the Langevin diffusion. Second, we provide an explicit representation of the optimal coefficients minimizing the asymptotic variance of the Langevin diffusion. Several examples of Bayesian inference problems demonstrate that the corresponding reduction in the variance is significant, and that in some cases it can be dramatic.
Open AccessDiffusion approximations and control variates for MCMC
A new methodology is presented for the construction of control variates to reduce the variance of additive functionals of Markov Chain Monte Carlo (MCMC) samplers. Our control variates are defined as linear combinations of functions whose coefficients are obtained by minimizing a proxy for the asymptotic variance. The construction is theoretically justified by two new results. We first show that the asymptotic variances of some well-known MCMC algorithms, including the Random Walk Metropolis and the (Metropolis) Unadjusted/Adjusted Langevin Algorithm, are close to the asymptotic variance of the Langevin diffusion. Second, we provide an explicit representation of the optimal coefficients minimizing the asymptotic variance of the Langevin diffusion. Several examples of Bayesian inference problems demonstrate that the corresponding reduction in the variance is significant, and that in some cases it can be dramatic.
- Open Access
The promises and pitfalls of Stochastic Gradient Langevin Dynamics
Stochastic Gradient Langevin Dynamics (SGLD) has emerged as a key MCMC algorithm for Bayesian learning from large scale datasets. While SGLD with decreasing step sizes converges weakly to the posterior distribution, the algorithm is often used with a constant step size in practice and has demonstrated successes in machine learning tasks. The current practice is to set the step size inversely proportional to $N$ where $N$ is the number of training samples. As $N$ becomes large, we show that the SGLD algorithm has an invariant probability measure which significantly departs from the target posterior and behaves like Stochastic Gradient Descent (SGD). This difference is inherently due to the high variance of the stochastic gradients. Several strategies have been suggested to reduce this effect; among them, SGLD Fixed Point (SGLDFP) uses carefully designed control variates to reduce the variance of the stochastic gradients. We show that SGLDFP gives approximate samples from the posterior distribution, with an accuracy comparable to the Langevin Monte Carlo (LMC) algorithm for a computational cost sublinear in the number of data points. We provide a detailed analysis of the Wasserstein distances between LMC, SGLD, SGLDFP and SGD and explicit expressions of the means and covariance matrices of their invariant distributions. Our findings are supported by limited numerical experiments.Open AccessThe promises and pitfalls of Stochastic Gradient Langevin Dynamics
Stochastic Gradient Langevin Dynamics (SGLD) has emerged as a key MCMC algorithm for Bayesian learning from large scale datasets. While SGLD with decreasing step sizes converges weakly to the posterior distribution, the algorithm is often used with a constant step size in practice and has demonstrated successes in machine learning tasks. The current practice is to set the step size inversely proportional to $N$ where $N$ is the number of training samples. As $N$ becomes large, we show that the SGLD algorithm has an invariant probability measure which significantly departs from the target posterior and behaves like Stochastic Gradient Descent (SGD). This difference is inherently due to the high variance of the stochastic gradients. Several strategies have been suggested to reduce this effect; among them, SGLD Fixed Point (SGLDFP) uses carefully designed control variates to reduce the variance of the stochastic gradients. We show that SGLDFP gives approximate samples from the posterior distribution, with an accuracy comparable to the Langevin Monte Carlo (LMC) algorithm for a computational cost sublinear in the number of data points. We provide a detailed analysis of the Wasserstein distances between LMC, SGLD, SGLDFP and SGD and explicit expressions of the means and covariance matrices of their invariant distributions. Our findings are supported by limited numerical experiments. - Open Access
Normalizing constants of log-concave densities
We derive explicit bounds for the computation of normalizing constants $Z$ for log-concave densities $\pi = \exp(-U)/Z$ with respect to the Lebesgue measure on $\mathbb{R}^d$. Our approach relies on a Gaussian annealing combined with recent and precise bounds on the Unadjusted Langevin Algorithm (High-dimensional Bayesian inference via the Unadjusted Langevin Algorithm, A. Durmus and E. Moulines). Polynomial bounds in the dimension $d$ are obtained with an exponent that depends on the assumptions made on $U$. The algorithm also provides a theoretically grounded choice of the annealing sequence of variances. A numerical experiment supports our findings. Results of independent interest on the mean squared error of the empirical average of locally Lipschitz functions are established.Open AccessNormalizing constants of log-concave densities
We derive explicit bounds for the computation of normalizing constants $Z$ for log-concave densities $\pi = \exp(-U)/Z$ with respect to the Lebesgue measure on $\mathbb{R}^d$. Our approach relies on a Gaussian annealing combined with recent and precise bounds on the Unadjusted Langevin Algorithm (High-dimensional Bayesian inference via the Unadjusted Langevin Algorithm, A. Durmus and E. Moulines). Polynomial bounds in the dimension $d$ are obtained with an exponent that depends on the assumptions made on $U$. The algorithm also provides a theoretically grounded choice of the annealing sequence of variances. A numerical experiment supports our findings. Results of independent interest on the mean squared error of the empirical average of locally Lipschitz functions are established. - Open Access
- 1 January 2018
- Published by Institute of Mathematical Statistics in Electronic Journal of Statistics
- Vol. 12 (1) , 851-889
Normalizing constants of log-concave densities
We derive explicit bounds for the computation of normalizing constants $Z$ for log-concave densities $\pi =\mathrm{e}^{-U}/Z$ w.r.t. the Lebesgue measure on $\mathbb{R}^{d}$. Our approach relies on a Gaussian annealing combined with recent and precise bounds on the Unadjusted Langevin Algorithm [15]. Polynomial bounds in the dimension $d$ are obtained with an exponent that depends on the assumptions made on $U$. The algorithm also provides a theoretically grounded choice of the annealing sequence of variances. A numerical experiment supports our findings. Results of independent interest on the mean squared error of the empirical average of locally Lipschitz functions are established.- 1 January 2018
- Published by Institute of Mathematical Statistics in Electronic Journal of Statistics
- Vol. 12 (1) , 851-889
Open AccessNormalizing constants of log-concave densities
We derive explicit bounds for the computation of normalizing constants $Z$ for log-concave densities $\pi =\mathrm{e}^{-U}/Z$ w.r.t. the Lebesgue measure on $\mathbb{R}^{d}$. Our approach relies on a Gaussian annealing combined with recent and precise bounds on the Unadjusted Langevin Algorithm [15]. Polynomial bounds in the dimension $d$ are obtained with an exponent that depends on the assumptions made on $U$. The algorithm also provides a theoretically grounded choice of the annealing sequence of variances. A numerical experiment supports our findings. Results of independent interest on the mean squared error of the empirical average of locally Lipschitz functions are established.