arXiv subjects

Carl-Erik Gauthier

Publications and source records attributed to Carl-Erik Gauthier.

9 recordsLinked to original sources

arXiv2025

Harnessing Mixed Features for Imbalance Data Oversampling: Application to Bank Customers Scoring

This study investigates rare event detection on tabular data within binary classification. Standard techniques to handle class imbalance include SMOTE, which generates synthetic samples from the minority class. However, SMOTE is intrinsically designed for continuous input variables. In fact, despite SMOTE-NC-its default extension to handle mixed features (continuous and categorical variables)-very few works propose procedures to synthesize mixed features. On the other hand, many real-world classification tasks, such as in banking sector, deal with mixed features, which have a significant impact on predictive performances. To this purpose, we introduce MGS-GRF, an oversampling strategy designed for mixed features. This method uses a kernel density estimator with locally estimated full-rank covariances to generate continuous features, while categorical ones are drawn from the original samples through a generalized random forest. Empirically, contrary to SMOTE-NC, we show that MGS-GRF exhibits two important properties: (i) the coherence i.e. the ability to only generate combinations of categorical features that are already present in the original dataset and (ii) association, i.e. the ability to preserve the dependence between continuous and categorical features. We also evaluate the predictive performances of LightGBM classifiers trained on data sets, augmented with synthetic samples from various strategies. Our comparison is performed on simulated and public real-world data sets, as well as on a private data set from a leading financial institution. We observe that synthetic procedures that have the properties of coherence and association display better predictive performances in terms of various predictive metrics (PR and ROC AUC...), with MGS-GRF being the best one. Furthermore, our method exhibits promising results for the private banking application, with development pipeline being compliant with regulatory constraints.

cs.LG↗

arXiv2021

Fermi acceleration in rotating drums

Consider hard balls in a bounded rotating drum. If there is no gravitation then there is no Fermi acceleration, i.e., the energy of the balls remains bounded forever. If there is gravitation, Fermi acceleration may arise. A number of explicit formulas for the system without gravitation are given. Some of these are based on an explicit realization, which we derive, of the well-known microcanonical ensemble measure.

math-ph↗

arXiv2019

Strongly self-interacting processes on the circle

The purpose of this paper is to investigate the long time behaviour for a self-interacting diffusion and a self-interacting velocity jump process. While the diffusion case has already been studied for some particular potential function, the second one, which belongs to the family of piecewise deterministic processes, is new. Depending on the underlying potential function's shape, we prove either the almost sure convergence or the recurrence for a natural extended process given by a change a variable.

math.PR↗

arXiv2018

Billiards with Markovian reflection laws

We construct a class of reflection laws for billiard processes in the unit interval whose stationary distribution for the billiard position and its velocity is the product of the uniform distribution and the standard normal distribution. These billiard processes have Markovian reflection laws, meaning their velocity is constant between reflections but changes in a Markovian way at reflection times.

math.PR↗

arXiv2018

Knudsen gas in flat tire

We consider random reflections (according to the Lambertian distribution) of a light ray in a thin variable width (but almost circular) tube. As the width of the tube goes to zero, properly rescaled angular component of the light ray position converges in distribution to a diffusion whose parameters (diffusivity and drift) are given explicitly in terms of the tube width.

math.PR↗

arXiv2017

Central Limit Theorem for a Self-Repelling Diffusion

We prove a Central Limit Theorem for the finite dimensional distributions of the displacement for the 1D self-repelling diffusion which solves \begin{equation*} dX_t =dB_t -\big(G'(X_t)+ \int_0^t F'(X_t-X_s)ds\big)dt, \end{equation*} where $B$ is a real valued standard Brownian motion and $F(x)=\sum_{k=1}^n a_k \cos(kx)$ with $n<\infty$ and $a_1,\cdots ,a_n >0$. In dimension $d\geq 3$, such a result has already been established by Horváth, Tóth and Vetö in \cite{HTV} in 2012 but not for $d=1,2$. Under an integrability condition, Tarrès, Tóth and Valkó conjectured in \cite{TTV} that a Central Limit Theorem result should also hold in dimension $d=1$.

math.PR↗

arXiv2016

Self-repelling diffusions on a Riemannian manifold

Let M be a compact connected oriented Riemannian manifold. The purpose of this paper is to investigate the long time behavior of a degenerate stochastic differential equation on the state space $M\times \mathbb{R}^{n}$; which is obtained via a natural change of variable from a self-repelling diffusion taking the form $$dX_{t}= σdB_{t}(X_t) -\int_{0}^{t}\nabla V_{X_s}(X_{t})dsdt,\qquad X_{0}=x$$ where $\{B_t\}$ is a Brownian vector field on $M$, $σ>0$ and $V_x(y) = V(x,y)$ is a diagonal Mercer kernel. We prove that the induced semi-group enjoys the strong Feller property and has a unique invariant probability $μ$ given as the product of the normalized Riemannian measure on M and a Gaussian measure on $\mathbb{R}^{n}$. We then prove an exponential decay to this invariant probability in $L^{2}(μ)$ and in total variation.

math.PR↗

arXiv2015

Self attracting diffusions on a sphere and application to a periodic case

This paper proves almost-sure convergence for the self-attracting diffusion on the unit sphere $$dX(t)=σdW_{t}(X(t))-a\int_{0}^{t}\nabla_{\mathbb{S}^n}V_{X_s}(X_t) dsdt,\qquad X(0)=x\in\mathbb{S}^n $$ %given by the stochastic differential equation: $$dX_{t}=σdW_{t}+a\int_{0}^{t}\sin(X_{t}-X_{s})dsdt, $$ where $σ>0$, $a < 0$, $V_y(x)=\langle x,y\rangle$ is the usual scalar product in $\mathbb{R}^n$, and $(W_{t}(.))_{t\geqslant 0}$ is a Brownian motion on $\mathbb{S}^n$. From this follows the almost-sure convergence of the real-valued self-attracting diffusion $$d\vartheta_{t}=σdW_{t}+a\int_{0}^{t}\sin(\vartheta_{t}-\vartheta_{s})dsdt, $$ where $(W_t)_{t\geqslant 0}$ is a real Brownian motion.

math.PR↗

arXiv2015

Self-repelling diffusions via an infinite dimensional approach

In the present work we study self-interacting diffusions following an infinite dimensional approach. First we prove existence and uniqueness of a solution with Markov property. Then we study the corresponding transition semigroup and, more precisely, we prove that it has Feller property and we give an explicit form of an invariant probability of the system.

math.PR↗