Scientific Publications of the University of Toulouse II Le Mirail
Not a member yet
92205 research outputs found
Sort by
Fourier Analysis on the Boolean Hypercube via Hoeffding Functional Decomposition
No full textFourier analysis on the Boolean hypercube is fundamentally defined as the orthogonal decomposition of the space of pseudo-Boolean functions with respect to the uniform probability measure. In this work, we propose an ANOVA-based generalization of the Fourier decomposition on the Boolean hypercube endowed with any arbitrary probability measure. We provide an \emph{explicit} decomposition basis which generalizes the Walsh-Hadamard (or parity functions) basis under any \emph{arbitrary} probability measure on the Boolean hypercube. We formulate the computation of the entire functional decomposition as a least squares problem and also provide a method to address the classical \emph{curse of dimensionality} challenge. We provide a comprehensive generalization of Fourier analysis on the Boolean hypercube, enabling the handling of non-uniform configuration spaces inherent to real-world machine learning tasks, \textit{e.g.} when dealing with \emph{one-hot encoded} features. Finally, we demonstrate its practical impact in the field of explainable AI, by conducting comparative studies with feature attribution methods such as SHAP or TreeHFD.</div
Part-night lighting fails to restore natural song rhythms in urban European Robins
No full textInternational audienceArtificial Light at Night (ALAN) disrupts the natural circadian rhythms of animals, often extending the activity of diurnal species into nocturnal periods. To mitigate ALAN's ecological impacts and reduce energy consumption, many municipalities have implemented Part-Night Lighting (PNL), which involves switching off streetlights during core nighttime hours. However, the effects of such temporal lighting reductions on animal activity patterns remain poorly understood. We investigate how PNL influences daily singing behavior in European Robins Erithacus rubecula during spring in an urban area of France, where streetlights are turned off between 11 p.m. and 6 a.m., creating brief illumination windows at dawn and dusk. We hypothesized that PNL would reduce the effects of ALAN on Robin vocal activity compared to Full-Night Lighting (FNL). Using passive acoustic monitoring, we recorded Robin song activity across three types of sites (unlit, PNL, and FNL) and four time periods (dawn, morning, afternoon, dusk). Robins sang significantly earlier at FNL and PNL sites than at unlit sites during dawn and morning, and later during dusk, with no significant differences between FNL and PNL. In the afternoon, song activity tended to be delayed at PNL sites compared to both FNL and unlit sites. Our findings indicate that even short illumination periods at dawn and dusk under PNL do not replicate unlit conditions. In an urban context, the similarity in vocal activity between PNL and FNL sites suggests that residual light from nearby lit areas and/or lingering effects of ALAN may continue to influence animal behavior under PNL regimes
Combining LLMs-based Conversational Agents and Ontologies for Open Data Research
No full textInternational audienceOpen Science has significantly increased the availability of heterogeneous scientific datasets. However, this abundance often results in difficulties identifying data that are truly relevant to a specific user needs, as pertinent resources may be obscured by large volumes of unrelated or poorly described data. This paper proposes an approach that combines semantically enriched metadata and LLM-based agents able to interpret natural language queries, in order to support the retrieval of appropriate datasets. This architecture enables the extraction and refinement of user intent and the generation of intelligible justifications for the retrieved results. To assess the performance of the proposed system, a comparative evaluation across multiple Earth Observation (EO) scenarios has been carried out. Four LLM agents has been evaluated -LLaMA 3.3 70B, Mistral Saba 24B, Deepseek-R1, and Qwen 32B -using metrics such as answer relevancy, contextual precision, recall, and faithfulness. The results, conducted with the Deepeval library associated with the LLaMA 3 8B model, show relatively high scores on answer relevancy and contextual precision, especially for the LLaMA and Deepseek-R1 models, demonstrating the benefits of LLM-based query interpretation and explanation generation
Microphone array geometry-independent multi-talker distant ASR: NTT system for DASR task of the CHiME-8 challenge
No full textInternational audienceIn this paper, we introduce a multi-talker distant automatic speech recognition (DASR) system we designed for the DASR task 1 of the CHiME-8 challenge. Our system performs speaker counting, diarization, and ASR. It handles a variety of recording conditions, from dinner parties to professional meetings and from two speakers to eight. We perform diarization first, followed by speech enhancement, and then ASR as the challenge baseline. However, we introduced several key refinements. First, we derived a powerful speaker diarization relying on end-to-end speaker diarization with vector clustering (EEND-VC), multi-channel speaker counting using enhanced embeddings from EEND-VC, and target-speaker voice activity detection (TS-VAD). For speech enhancement, we introduced a novel microphone selection rule to better select the most relevant microphones among those distributed microphones and investigated improvements to beamforming. Finally, for ASR, we developed several models exploiting Whisper and WavLM speech foundation models. In this paper, we present the original results we submitted to the challenge and updated results we obtained afterward. Our strongest system achieves a 63% relative macro tcpWER improvement over the baseline and outperforms the challenge best results on the NOTSOFAR-1 meeting evaluation data among geometry-independent systems
Dosimetry and preclinical evaluation of long-term radiotoxicity following treatment with 212Pb alpha-radioimmunotherapy targeting CD20
No full textInternational audienceBackground: Radioimmunotherapy (RIT) with α-emitters represents an attractive alternative for the treatment of refractory Non-Hodgkin lymphoma (NHL) due to the high linear energy transfer and short path length of α-radiation in tissues. We have previously shown that α-RIT with [212Pb]Pb-TCMC-rituximab is potentially useful for treatment of NHL. In this study, we performed radiation dosimetry and evaluated the long-term toxicity in mice to determine safety of [212Pb]Pb-TCMC-rituximab,.Results: Biodistribution data obtained after intravenous administration of [212Pb]Pb-TCMC-rituximab (185 kBq) in healthy mice were used to calculate the absorbed radiation doses from [212Pb]Pb-TCMC-rituximab. Analyses show that the alveolar-interstitial, kidneys, and spleen receive the highest dose. In order to evaluate the toxicity of RIT for up to 9 months, [212Pb]Pb-TCMC-rituximab was administered intravenously in healthy C57BL/6 mice (277.5 and 555 kBq) and in a B-NHL immunocompetent mouse model (277.5 kBq, specific activity of 37 or 370 MBq/mg). Our previous study revealed a high efficacy of [212Pb]Pb-TCMC-rituximab at 277.5 kBq and that activities of 185-370 kBq of [212Pb]Pb-TCMC-rituximab were well-tolerated. However, in this long-term study, toxicity emerged in healthy mice after four months. The median survival for the 277.5 and 555 kBq groups were 189 and 161 days, respectively. There was no significant hepatic toxicity, but there was a significant increase in urea and creatinine levels at 6 months, indicating long-term renal toxicity (p < 0.001). These results were supported by histopathological data. Long-term renal toxicity is also observed in the toxicity study performed on tumor model with two specific activities of [212Pb]Pb-TCMC-rituximab. Nevertheless, this toxicity was reduced at 370 MBq/mg compared to 37 MBq/mg.Conclusion: This study shows that long-term toxicity is induced by [212Pb]Pb-TCMC-rituximab, particularly affecting the kidneys. However, it highlights that this renal toxicity can be reduced through optimization, possibly by modifying the specific activity of the treatment
Attitudes et perspectives des professionnels des centres français de DPI concernant le dépistage des aneuploïdies et l’utilisation du NGS en DPI
No full textNational audienceIntroduction : Le diagnostic préimplantatoire pour aneuploïdies (DPI-A) vise à détecter les anomalies chromosomiques embryonnaires avant transfert. Parallèlement, le séquençage à haut débit (NGS) peut être appliqué au DPI, notamment en remplacement de l’analyse par microsatellites pour le diagnostic indirect des maladies monogéniques. En France, le DPI-A, bien que discuté lors des dernières révisions de la loi de bioéthique, n’a pas été retenu et reste interdit. Le NGS n’est pas utilisé en routine. Cette étude, mandatée par la FFGH, explore les connaissances et perceptions des professionnels français du DPI concernant ces techniques. Méthodologie : Une quarantaine d’entretiens semi-directifs ont été menés auprès des différents professionnels des cinq centres de DPI français. Les entretiens ont été analysés de façon thématique, transversale puis comparative. Résultats : Les perceptions du DPI-A divergent. Certains y voient un outil d’optimisation de l’assistance médicale à la procréation (AMP), d’autres rappellent des données bibliographiques contradictoires et privilégient sa finalité de prévention des aneuploïdies. S’il était autorisé, la majorité y serait favorable, avec des critères d’inclusion variables : couples déjà en DPI, ou aussi certaines FIV (âge maternel élevé, fausses couches répétées, échecs d’implantation). Aucun n’envisage son usage hors indication d’AMP. Les risques cités concernent l’exclusion d’embryons viables ainsi que l’interprétation complexe des mosaïques. Le NGS en DPI reste méconnu hormis des généticiens. Ses avantages perçus sont la réduction du temps de mise au point, l’usage d’une méthode unique et l’ouverture à de nouvelles indications. Les inquiétudes portent sur la découverte de données incidentes ou de variants de signification incertaine, la perte de compétences humaines, ainsi que la sélection excessive d’embryons, pouvant réduire les chances de grossesse et soulevant des enjeux éthiques, dont la crainte d’une forme d’ « eugénisme ». Le frein principal est la nécessité d’outils bio-informatiques pour masquer les régions non ciblées. L’autorisation de ces pratiques nécessiterait des adaptations organisationnelles : passage de la biopsie de J3 à J5 avec congélation systématique, et, pour le DPI-A, hausse probable des demandes impliquant davantage de moyens humains et matériels, ou une centralisation nationale. Des différences inter-centres et interprofessions sont relevées selon les techniques utilisées, avec un sentiment de retard vis-à-vis de l’étranger, et une influence notable des expériences personnelles sur les positions exprimées. L’aspect financier était spontanément peu évoqué. Conclusion : Cette enquête auprès des professionnels français du DPI révèle attentes et réserves vis-à-vis du DPI-A et du NGS. La majorité se dit favorable à leur introduction, tout en soulignant des freins éthiques, techniques et organisationnels. Une réflexion collective apparaît nécessaire pour encadrer leur éventuelle mise en œuvre
Embryon, fœtus: Quels statuts juridiques ?
No full texthttps://sms.hypotheses.org/57259Debates around abortion and embryonic research continually raise the same question: what status should be granted to the being developing before birth? Is it merely a cluster of cells? Or should it instead be regarded as a full person from the moment of conception? These positions clash in public debates and appear irreconcilable. How can French society reconcile two apparently contradictory movements: women’s autonomy and the preservation of embryonic and foetal life?Les débats autour de l’avortement et de la recherche sur l’embryon soulèvent sans cesse la même question : quel statut accorder à l’être qui se développe avant la naissance ? S’agit-il d’un simple amas de cellules ? Doit-on au contraire le considérer comme une personne à part entière dès la conception ? Les positions s’affrontent dans les débats et semblent inconciliables. Comment la société française peut-elle concilier deux mouvements apparemment contradictoires : l’autonomie des femmes et la préservation de la vie embryonnaire et fœtale
ROI-NeRFs: Hi-Fi Visualization of Objects of Interest within a Scene by NeRFs Composition
No full textInternational audienceEfficient and accurate 3D reconstruction is crucial for cultural heritage applications. This study addresses the challenge of visualizing objects in complex scenes at high levels of detail (LOD) using Neural Radiance Fields (NeRFs), improving visual fidelity for selected objects while maintaining computational efficiency. The proposed ROI-NeRFs framework divides the scene into a Scene NeRF, capturing the overall scene at moderate detail, and multiple Region Of Interest (ROI) NeRFs, focusing on user-defined objects. An object-focused camera selection module automatically groups relevant cameras for each NeRF during the decomposition phase. In contrast, a Ray-level Compositional Rendering technique in the composition phase integrates Scene and ROI NeRFs for simultaneous multi-object rendering. Experiments on two real-world datasets, including a complex eighteenth-century cultural heritage room, demonstrate superior performance over baseline methods, enhancing LOD in object regions, minimizing artifacts, and maintaining efficient inference. INTRODUCTION3D reconstruction of cultural heritage sites is essential for creating digital twins used in archiving, conservation [Kong and Hucks, 2023], archaeology [Haibt, 2024], and interactive museum exhibitions [Liu and Chang, 2024].Neural Radiance Field (NeRF) [Mildenhall et al., 2021] has emerged as a powerful alternative for novel view synthesis (NVS) with unprecedented image quality. NeRF learns scenes as continuous fields of volume density and view-dependent color using MLPs, achieving photorealistic rendering through differentiable volume rendering. However, most NeRF-based methods train a single model on all images of a scene, yielding renderings with a uniform level of detail (LOD). In large-scale scenes, this fixed resolution might lead to poor-quality when viewing close-up specific objects or regions of interest (ROI), while increasing detail globally would incur prohibitive computational costs.</div
PCA of Probability Measures: Sparse and Dense Sampling Regimes
No full textA common approach to perform PCA on probability measures is to embed them into a Hilbert space where standard functional PCA techniques apply. While convergence rates for estimating the embedding of a single measure from m samples are well understood, the literature has not addressed the setting involving multiple measures. In this paper, we study PCA in a double asymptotic regime where n probability measures are observed, each through m samples. We derive convergence rates of the form n -1/2 + m -α for the empirical covariance operator and the PCA excess risk, where α > 0 depends on the chosen embedding. This characterizes the relationship between the number n of measures and the number m of samples per measure, revealing a sparse (small m) to dense (large m) transition in the convergence behavior. Moreover, we prove that the dense-regime rate is minimax optimal for the empirical covariance error. Our numerical experiments validate these theoretical rates and demonstrate that appropriate subsampling preserves PCA accuracy while reducing computational cost.</div
The Programme for International Student Assessment (PISA): a scientific (counter)model for research in comparative education?
No full textInternational audienceSince the launch of PISA in 2000, the success of this OECD-led evaluation has been particularly remarkable. In fact, PISA does not shy away from introducing itself in its reports as a leading reference in comparative education. This chapter analyses precisely the issues raised by the adoption of the PISA methodology as research model in comparative education. Three major issues will be analysed to describe how the PISA publications stray from the common scientific norms: from the perspective of taking into account the critical studies that has been carried out on PISA; from the perspective of the shift from analyzing relationships between variables and results to causal links; from the perspective of a more or less implicit reference to a globalising or even neoliberal political understanding. Through a critical analysis of these issues raised by PISA, this chapter will highlight the importance of reflexivity in comparative education to support other approaches to education and research in the face of a dominant discourse on a worldwide scale