arXiv subjects

Guénaël Cabanes

Publications and source records attributed to Guénaël Cabanes.

4 recordsLinked to original sources

arXiv2026

MILE-RefHumEval: A Reference-Free, Multi-Independent LLM Framework for Human-Aligned Evaluation

We introduce MILE-RefHumEval, a reference-free framework for evaluating Large Language Models (LLMs) without ground-truth annotations or evaluator coordination. It leverages an ensemble of independently prompted evaluators guided by a human-aligned schema, supporting both discrete and continuous scoring judgement. With task-specific prompts from best candidate selection, summarization and image captioning to dialogue, MILE-RefHumEval provides flexible, interpretable, and scalable assessments. Experiments show it aligns closely with human judgments, outperforms prior methods, and reduces computational overhead, offering an efficient, robust, and human-aligned solution for real-world LLM evaluation.

cs.CL↗

arXiv2025

An optimal baseline selection methodology for data-driven damage detection and temperature compensation in acousto-ultrasonics

The global trends in the construction of modern structures require the integration of sensors together with data recording and analysis modules so that their integrity can be continuously monitored for safe-life, economic and ecological reasons. This process of measuring and analyzing the data from a distributed sensor network all over a structural system in order to quantify its condition is known as structural health monitoring (SHM). Guided ultrasonic wave-based techniques are increasingly being adapted and used in several SHM systems which benefit from built-in transduction, large inspection ranges, and high sensitivity to small flaws. Nonetheless, for the design of a trustworthy health monitoring system, a vast amount of information regarding the inherent physical characteristics of the sources and their propagation and interaction across the structure is crucial. Moreover, any SHM system which is expected to transition to field operation must take into account the influence of environmental and operational changes which cause modifications in the stiffness and damping of the structure and consequently modify its dynamic behaviour. On that account, special attention is paid in this paper to the development of an efficient SHM methodology where robust signal processing and pattern recognition techniques are integrated for the correct interpretation of complex ultrasonic waves within the context of damage detection and identification. The methodology is based on an acousto-ultrasonics technique where the discrete wavelet transform is evaluated for feature extraction and selection, linear principal component analysis for data-driven modelling and selforganizing maps for a two-level clustering under the principle of local density. At the end, the methodology is experimentally demonstrated and results show that all the damages were detectable and identifiable.

eess.SP↗

arXiv2023

Deep Learning Approaches for Dynamic Mechanical Analysis of Viscoelastic Fiber Composites

The increased adoption of reinforced polymer (RP) composite materials, driven by eco-design standards, calls for a fine balance between lightness, stiffness, and effective vibration control. These materials are integral to enhancing comfort, safety, and energy efficiency. Dynamic Mechanical Analysis (DMA) characterizes viscoelastic behavior, yet there's a growing interest in using Machine Learning (ML) to expedite the design and understanding of microstructures. In this paper we aim to map microstructures to their mechanical properties using deep neural networks, speeding up the process and allowing for the generation of microstructures from desired properties.

cs.LG↗

arXiv2021

On the use of Wasserstein metric in topological clustering of distributional data

This paper deals with a clustering algorithm for histogram data based on a Self-Organizing Map (SOM) learning. It combines a dimension reduction by SOM and the clustering of the data in a reduced space. Related to the kind of data, a suitable dissimilarity measure between distributions is introduced: the $L_2$ Wasserstein distance. Moreover, the number of clusters is not fixed in advance but it is automatically found according to a local data density estimation in the original space. Applications on synthetic and real data sets corroborate the proposed strategy.

cs.LG↗