Institution profile

University of Montpellier 2

Academic institutioneurope · fr
Official website
Research library63linked papers
Opportunities0open roles
Selected work

Representative Papers

Approximating branchwidth on parametric extensions of planarity

Apr 10, 2023International Workshop on Graph-Theoretic Concepts in Computer Science

This paper addresses the branchwidth approximation problem for graph classes excluding two fixed graphs $H_1$ and $H_2$, each embeddable on the torus or the projective plane. For this broad family of non-planar graphs, we extend the Seymour–Thomas Ratcatcher algorithm—previously applicable only to planar graphs—to handle toroidal and projective-planar forbidden minors. Our method integrates the Graph Minor Structure Theorem, extraction of planar subgraphs, and constructive tree decomposition. The resulting algorithm runs in $O(|V|^3)$ time and achieves a constant additive approximation guarantee: the error depends solely on $H_1$ and $H_2$, not on the input graph size. This work overcomes a fundamental bottleneck—the intractability of exact branchwidth computation beyond planar graphs—and provides the first polynomial-time constant-additive approximation algorithm for branchwidth with rigorous theoretical guarantees on a wide class of non-planar graphs.

2 citationsRead paper

Obstructions for Minor-Closed Classes of limiting Densities Below 3/2

Jun 23, 2026

This work investigates the limit densities of minor-closed graph classes, with a focus on those whose densities are strictly less than $3/2$. By integrating tools from extremal graph theory, closure properties under graph minors, and combinatorial enumeration, the paper provides the first complete characterization of all minimal minor-closed graph classes whose limit densities lie in the interval $[0, 3/2)$. It further establishes that the corresponding minimal forbidden minor sets for these classes are finite. Building on this structural characterization, the authors develop an algorithmic framework capable of deciding, in time $2^{\text{poly}(n)}$, whether the limit density of a given minor-closed graph class is below $3/2$, thereby resolving the decidability question for this critical density threshold.

0 citationsRead paper

Evaluating the Interpretability of Sparse Autoencoders with Concept Annotations

Jun 23, 2026

This work addresses the lack of quantitative methods for evaluating semantic alignment between latent variables in sparse autoencoders (SAEs) and human-interpretable concepts. To this end, the authors propose a human-grounded evaluation framework that obviates user studies by leveraging synthetic datasets (synCUB and synCOCO), a many-to-one matching algorithm termed Fully-Binary Matching Pursuit (FBMP), and an alignment metric called TAPAScore based on attribute perturbations. The framework systematically quantifies correspondence between SAE features and human-annotated concepts. Experimental results demonstrate that SAEs with moderate dictionary sizes achieve optimal interpretability, while excessive overcompleteness degrades perturbation-based alignment. Furthermore, the proposed method reliably distinguishes between trained and untrained SAEs, establishing a robust quantitative benchmark for interpretability assessment.

0 citationsRead paper
Recent publications

Latest Papers

Obstructions for Minor-Closed Classes of limiting Densities Below 3/2

Jun 23, 2026

This work investigates the limit densities of minor-closed graph classes, with a focus on those whose densities are strictly less than $3/2$. By integrating tools from extremal graph theory, closure properties under graph minors, and combinatorial enumeration, the paper provides the first complete characterization of all minimal minor-closed graph classes whose limit densities lie in the interval $[0, 3/2)$. It further establishes that the corresponding minimal forbidden minor sets for these classes are finite. Building on this structural characterization, the authors develop an algorithmic framework capable of deciding, in time $2^{\text{poly}(n)}$, whether the limit density of a given minor-closed graph class is below $3/2$, thereby resolving the decidability question for this critical density threshold.

0 citationsRead paper

Evaluating the Interpretability of Sparse Autoencoders with Concept Annotations

Jun 23, 2026

This work addresses the lack of quantitative methods for evaluating semantic alignment between latent variables in sparse autoencoders (SAEs) and human-interpretable concepts. To this end, the authors propose a human-grounded evaluation framework that obviates user studies by leveraging synthetic datasets (synCUB and synCOCO), a many-to-one matching algorithm termed Fully-Binary Matching Pursuit (FBMP), and an alignment metric called TAPAScore based on attribute perturbations. The framework systematically quantifies correspondence between SAE features and human-annotated concepts. Experimental results demonstrate that SAEs with moderate dictionary sizes achieve optimal interpretability, while excessive overcompleteness degrades perturbation-based alignment. Furthermore, the proposed method reliably distinguishes between trained and untrained SAEs, establishing a robust quantitative benchmark for interpretability assessment.

0 citationsRead paper

Detecting Pen-In-Air States from Video: A Proof-of-Concept Toward Complementary Handwriting Analysis

Jun 01, 2026

This study addresses the limitation of conventional digitizing tablets in capturing pen-lift (pen-up) movements during handwriting, which hinders comprehensive assessment of writing impairments such as dysgraphia. To overcome this, the authors propose a non-invasive approach that infers pen-tip contact status solely from top-view video recordings. The method integrates YOLO-based pen-tip tracking, extracts kinematic features, and employs a machine learning classifier for frame-level pen-up detection. A novel dataset with manual annotations is introduced, and the work demonstrates for the first time that top-view video can serve as a low-cost, unobtrusive complement to digitizing tablets for detecting in-air writing gestures. Under leave-one-video-out cross-validation, the system achieves an F₂ score of 0.805 for pen-up event detection, indicating high recall reliability suitable for screening applications.

0 citationsRead paper