Publications

2026

• Multimodal Cultural Heritage Knowledge Graph Extension with Language and Vision Models
  
  • Zhang Yang
  • Mimouni Nada
  • Moissinac Jean-Claude
  • Hamdi Fayçal
  Journal on Computing and Cultural Heritage, Association for Computing Machinery, 2026. The preservation and interpretation of cultural heritage increasingly rely on digital technologies, among which Knowledge Graphs (KGs) stand out for their ability to structure vast amounts of data. However, the construction and expansion of these KGs often face challenges due to the diverse and complex nature of cultural heritage information. In this paper, we propose a novel approach for extending KG resources in the domain of cultural heritage, which we applied to French data. First, we introduce a new knowledge graph in the domain of French cultural heritage, WJoconde, which is distinguished by its multimodality as it integrates both textual and image information of the entities. We further introduce three variants of WJoconde to facilitate downstream research, such as Knowledge Graph Completion (KGC). We also built a comprehensive benchmark for KGC methods on our dataset. Second, we propose a new framework for extending cultural heritage KGs using multi-modal approaches leveraging Large Language Models (LLMs) and Vision-Language Models (VLMs), which includes automated data extraction from unstructured resources combined with a special validation pipeline for grounding the output of both models, to further extend WJoconde. Our results show that by integrating the rich text and image information in cultural heritage data, we can efficiently enhance KGs with high reliability. We open-source all code and benchmark datasets with text and images, as well as the original data with an interactive access point (10.48550/arXiv.2605.17669)
  DOI : 10.48550/arXiv.2605.17669
• A Decade of Software Reproducibility in the Nix Package Ecosystem
  
  • Malka Julien
  • Zacchiroli Stefano
  • Zimmermann Théo
  Empirical Software Engineering, Springer Verlag, 2026. <div><p>We report a large-scale empirical study of two aspects of software reproducibility-rebuildability and bitwise reproducibility-in the Nix package ecosystem. Using 29 evenly spaced historical snapshots over a decade of history of the nixpkgs repository (2015-2024) we attempted to rebuild and bitwise-compare the build outputs of tens of thousands of packages per snapshot. Our experiment produced a dataset of build metadata and logs for 1 321 000 package builds and preserved 166 523 diffoscopes for unreproducible outputs.</p><p>We find that functional package management enables extremely high rebuildability over time (near-universal ability to reconstitute historical build environments and rebuild software packages), while bitwise reproducibility has steadily improved and reaches a high point in recent years (up to 93% in 2024). Early years show substantially lower bitwise reproducibility, indicating that functional package management alone does not guarantee bitwiseidentical outputs, and that the observed high level of bitwise reproducibility is not solely due to the package management approach. Common causes of unreproducibility, both in the rebuildability and bitwise reproducibility dimensions, include management of dates in build and test processes; we quantify their prevalence and other common causes using manual analysis of logs of rebuild failures and automated analysis of diffoscopes.</p></div> (10.1007/s10664-026-10924-1)
  DOI : 10.1007/s10664-026-10924-1
• Of All StrIPEs: Investigating Structure-informed Positional Encoding for Efficient Music Generation
  
  • Agarwal Manvi
  • Wang Changhong
  • Richard Gaël
  IEEE Transactions on Audio, Speech and Language Processing, Institute of Electrical and Electronics Engineers, 2026, 34, pp.1388-1400. While music remains a challenging domain for generative models like Transformers, a two-pronged approach has recently proved successful: inserting musically-relevant structural information into the positional encoding (PE) module and using kernel approximation techniques based on Random Fourier Features (RFF) to lower the computational cost from quadratic to linear. Yet, it is not clear how such RFF-based efficient PEs compare with those based on rotation matrices, such as Rotary Positional Encoding (RoPE). In this paper, we present a unified framework based on kernel methods to analyze both families of efficient PEs. We use this framework to develop a novel PE method called RoPEPool, capable of extracting causal relationships from temporal sequences. Using RFF-based PEs and rotation-based PEs, we demonstrate how seemingly disparate PEs can be jointly studied by considering the interactions they induce between two descriptive levels of the data: the input, capturing quickly-varying components, and the prior, capturing slowly-varying components. For empirical validation, we use a symbolic music generation task, namely, melody harmonization. We show that RoPEPool, combined with highly-informative structural priors, outperforms all methods. (10.1109/TASLPRO.2026.3662483)
  DOI : 10.1109/TASLPRO.2026.3662483
• Generalised contextuality of continuous variable quantum theory can be revealed with a single projective measurement
  
  • Jokinen Pauli
  • Weilenmann Mirjam
  • Plávala Martin
  • Pellonpää Juha-Pekka
  • Kiukas Jukka
  • Uola Roope
  arxiv.org, 2026. Generalized contextuality is a possible indicator of non-classical behaviour in quantum information theory. In finite-dimensional systems, this is justified by the fact that noncontextual theories can be embedded into some simplex, i.e. into a classical theory. We show that a direct application of the standard definition of generalized contextuality to continuous variable systems does not envelope the statistics of some basic measurements, such as the position observable. In other words, we construct families of fully classical, i.e. commuting, measurements that nevertheless can be used to show contextuality of quantum theory. To overcome the apparent disagreement between the two notions of classicality, that is commutativity and noncontextuality, we propose a modified definition of generalised contextuality for continuous-variable systems. The modified definition is based on a physically-motivated approximation procedure, that uses only finite sets of measurement effects. We prove that in the limiting case this definition corresponds exactly to an extension of noncontextual models that benefits from non-constructive response functions. In the process, we discuss the extension of a known connection between contextuality and no-broadcasting to the continuous-variable scenario, and prove structural results regarding fixed points of infinite-dimensional entanglement breaking channels. (10.48550/arXiv.2601.14067)
  DOI : 10.48550/arXiv.2601.14067
• Rate of convergence of the conditioned random walk towards the Brownian bridge
  
  • Decreusefond Laurent
  • Jacquet Antonin
  , 2026. <div><p>We study the rate of convergence of two discrete processes towards the Brownian bridge: the random walk conditioned to be zero at time 2n and the empirical process which appears in the Glivencko-Cantelli theorem. Combining a functional Stein method with a Radon-Nikodym representation of the bridge, we bound the Fortet-Mourier distance between these conditioned processes and the Brownian bridge.</p></div>
• Distilling Learned Image Compression Models: An Analytical Approach for Low-Latency FPGA Deployment
  
  • Mazouz Alaa Eddine
  • Chaudhuri Sumanta
  • Cagnazzo Marco
  • Mitrea Mihai
  • Tartaglione Enzo
  • Zatt Bruno
  • Fiandrotti Attilio
  IEEE Transactions on Multimedia, Institute of Electrical and Electronics Engineers, 2026. <div><p>Learned Image Compression (LIC) models now rival traditional video codecs in rate-distortion (RD) efficiency, spurring interest in hardware-friendly deployments. However, most existing implementations take an LIC model and fit it to a specific hardware platform through time-consuming, lowlevel hardware optimizations. Moreover, these methods are not designed to meet a pre-established target latency, leading to suboptimal complexity-efficiency-latency trade-offs. We propose a paradigm for distilling a student LIC model under a target latency constraint, avoiding the need for low-level hardware redesign. First, we establish an analytical relationship between the number of convolutional channels of a LIC model and its latency. Second, we introduce a framework to distill a large LIC model into a latency-bound student. Finally, we design a pipelined FPGA architecture that employs mixed-precision quantization, parallel processing, and optimized resource allocation for maximum efficiency. RD efficiency is preserved thanks to a hardwarefriendly GDN/iGDN module end-to-end integrated within our LIC pipeline. Experiments on a ZCU102 FPGA show that our approach achieves competitive RD efficiency with explicit latency control, reaching up to 60 fps for HD content while consuming less than 2.14 J/frame.</p></div>
• Statistically Robust Resource Block Allocation for Satellite Communications
  
  • Manapragada Chaitanya
  • Decreusefond Laurent
  • Martins Philippe
  , 2026. It is critical to dimension (accurately estimate capacity of) a satellite system prior to deployment, as it is very expensive to reconfigure launched satellite systems that fail to meet demand or that waste capacity. The fundamental requirement is a dimensioning rule for resource blocks (RBs) given a satellite footprint and a target overload probability (target Quality-of-Service). The rule must be robust to the spatial covariance structure of signal attenuation, which is generally unknown both at the time of pre-deployment dimensioning and afterwards. Existing approaches address parts of this problem, but there does not yet exist a footprint-level RB dimensioning rule for the satellite context. We develop such a rule: starting with a Gaussian attenuation field that induces a covariance structure inspired by classical work on spatial covariance of attenuation, we sample users at random along with their field-based attenuation values, and estimate aggregate RB demand for a target overload probability. We do this in two complementary ways: a Monte Carlo route that gives a simulation-derived RB budget for a given target overload probability, and a concentration route that gives a conservative analytic upper bound on the target overload probability for a given RB budget (such as the one obtained through simulation). Taken together, these complementary approaches give a principled way to dimension RBs for a satellite footprint under spatially correlated attenuation.
• UNSUPERVISED DOMAIN ADAPTATION WITH TARGET-ONLY MARGIN DISPARITY DISCREPANCY
  
  • Miralles Gauthier
  • Le Folgoc Loic
  • Jugnon Vincent
  • Gori Pietro
  , 2026. <div><p>In interventional radiology, Cone-Beam Computed Tomography (CBCT) is a helpful imaging modality that provides guidance to practicians during minimally invasive procedures. CBCT differs from traditional Computed Tomography (CT) due to its limited reconstructed field of view, specific artefacts, and the intra-arterial administration of contrast medium. While CT benefits from abundant publicly available annotated datasets, interventional CBCT data remain scarce and largely unannotated, with existing datasets focused primarily on radiotherapy applications. To address this limitation, we leverage a proprietary collection of unannotated interventional CBCT scans in conjunction with annotated CT data, employing domain adaptation techniques to bridge the modality gap and enhance liver segmentation performance on CBCT. We propose a novel unsupervised domain adaptation (UDA) framework based on the formalism of Margin Disparity Discrepancy (MDD), which improves target domain performance through a reformulation of the original MDD optimization framework. Experimental results on CT and CBCT datasets for liver segmentation demonstrate that our method achieves state-of-the-art performance in UDA, as well as in the few-shot setting.</p></div>
• Is Phase Really Needed for Weakly-Supervised Dereverberation ?
  
  • Rodrigues Marius
  • Bahrman Louis
  • Badeau Roland
  • Richard Gaël
  , 2026. In unsupervised or weakly-supervised approaches for speech dereverberation, the target clean (dry) signals are considered to be unknown during training. In that context, evaluating to what extent information can be retrieved from the sole knowledge of reverberant (wet) speech becomes critical. This work investigates the role of the reverberant (wet) phase in the time-frequency domain. Based on Statistical Wave Field Theory, we show that late reverberation perturbs phase components with white, uniformly distributed noise, except at low frequencies. Consequently, the wet phase carries limited useful information and is not essential for weakly supervised dereverberation. To validate this finding, we train dereverberation models under a recent weak supervision framework and demonstrate that performance can be significantly improved by excluding the reverberant phase from the loss function.
• Applying Morphological Operations to Subsets of Points for the Discovery of Repeated Musical Patterns and Their Variations
  
  • Lascabettes Paul
  • Quaetaert Nils
  • Daniel Mathys
  • Andreatta Moreno
  • Bloch Isabelle
  Journal of Mathematical Imaging and Vision, Springer Verlag, 2026. This article deals with the discovery of repeated patterns and their variations in a discrete representation of musical data. This task consists in identifying repetitions within a set of points in R2 , where each point represents a musical note whose coordinates are its onset and its pitch value. A common approach is to compute all the possible translations between points in order to discover repeated musical patterns. In this paper, we propose to start from specific subsets of points and to complete them by using morphological operations to form repeated patterns. Moreover, these operations can be extended to discover pattern variations given a particular approximation. This method not only reveals certain variations of the given subset of points, but also adds specific points to it despite the fact that they were not initially present. We apply our approach to the collection of 24 fugues from the first book of Bach's Well-Tempered Clavier. In this particular case, we consider the first m points as the subset to be completed by the morphological operators. We demonstrate that specific values of m enable the discovery of the subject and its occurrences, whereas the smallest values identify truncated versions of it. We compare our approach with previous work on the analysis of Bach's fugues and illustrate the results for different values of m with graphical representations including both exact and approximate repetitions. (10.1007/s10851-026-01301-0)
  DOI : 10.1007/s10851-026-01301-0
• Convergence rate for the coupon collector's problem with Stein's method
  
  • Costacèque Bruno
  • Decreusefond Laurent
  Stochastic Processes and their Applications, Elsevier, 2026, 193, pp.104835. The functional characterization of a measure, an essential but delicate aspect of Stein's method, is shown to be accessible for stable probability distributions on convex cones. This notion encompasses the usual stable distributions \textit{e.g.} Gaussian, Pareto, \textit{etc.} but also the max-stable distributions: Weibull, Gumbel and Fréchet. We use the definition of max-stability to define a Markov process whose invariant measure is the stable measure of interest. In this paper, we focus on the Gumbel distribution and show how this construction can be applied to estimate the rate of convergence in the classical coupon collector's problem. (10.1016/j.spa.2025.104835)
  DOI : 10.1016/j.spa.2025.104835
• Contrastive Learning under Noisy Temporal Self-Supervision for Colonoscopy Videos
  
  • Parolari Luca
  • Gori Pietro
  • Ballan Lamberto
  • Biffi Carlo
  • Folgoc Loic Le
  , 2026. Learning robust representations of polyp tracklets is key to enabling multiple AI-assisted colonoscopy applications, from polyp characterization to automated reporting and retrieval. Supervised contrastive learning is an effective approach for learning such representations, but it typically relies on correct positive and negative definitions. Collecting these labels requires linking tracklets that depict the same underlying polyp entity throughout the video, which is costly and demands specialized clinical expertise. In this work, we leverage the sequential workflow of colonoscopy procedures to derive self-supervised associations from temporal structure. Since temporally derived associations are not guaranteed to be correct, we introduce a noise-aware contrastive loss to account for noisy associations. We demonstrate the effectiveness of the learned representations across multiple downstream tasks, including polyp retrieval and re-identification, size estimation, and histology classification. Our method outperforms prior self-supervised and supervised baselines, and matches or exceeds recent foundation models across all tasks, using a lightweight encoder trained on only 27 videos. Code is available at github.com/lparolari/ntssl.
• Agentic Much? Adoption of Coding Agents on GitHub
  
  • Robbes Romain
  • Matricon Théo
  • Degueule Thomas
  • Hora Andre
  • Zacchiroli Stefano
  ACM Transactions on Software Engineering and Methodology, Association for Computing Machinery, 2026, pp.1-43. In the first half of 2025, coding agents have emerged as a category of development tools that have very quickly transitioned to the practice. Unlike ''traditional'' code completion LLMs such as Copilot, agents like Cursor, Claude Code, or Codex operate with high degrees of autonomy, up to generating complete pull requests starting from a developer-provided task description. This new mode of operation is poised to change the landscape in an even larger way than code completion LLMs did, making the need to study their impact critical. Also, unlike traditional LLMs, coding agents tend to leave more explicit traces in software engineering artifacts, such as co-authoring commits or pull requests. We leverage these traces to present the first large-scale study (128,018 projects) of the adoption of coding agents on GitHub, finding an estimated adoption rate of 22.20%--28.66%, which is very high for a technology only a few months old--and increasing. We carry out an in-depth study of the adopters we identified, finding that adoption is broad: it spans the entire spectrum of project maturity; it includes established organizations; and it concerns diverse programming languages or project topics. At the commit level, we find that commits assisted by coding agents are larger than commits only authored by human developers, and have a large proportion of features and bug fixes. These findings highlight the need for further investigation into the practical use of coding agents. (10.1145/3822180)
  DOI : 10.1145/3822180
• Event Detection and Localization Using a Multiple-Input-Multiple-Output Distributed Fiber Sensor with Birefringence and Phase Estimation
  
  • Prato Diane
  • Sheramin Mehran Mokthari
  • Gabet Renaud
  • Awwad Élie
  , 2026, pp.17. We present a numerical modeling approach for a Multiple-Input-Multiple-Output Distributed Acoustic Sensing (MIMO-DAS) system, incorporating both phase and polarization dynamics. We demonstrate the ability of the proposed architecture to provide a distributed estimation of the fiber effective linear birefringence magnitude and of the optical phase of the backscattered signal to detect and localize dynamic events with a mean spatial resolution of 1.3m. This allows for increased sensitivity to disturbances that act transversely on the fiber, since the estimated fiber effective birefringence magnitude will be responsive to perturbations that break cylindrical symmetry (anisotropic transverse strains), while the phase common to both polarization tributaries will show great sensitivity to pure longitudinal strains. Analyzing these two quantities is therefore of use to discriminate between purely axisymmetric strains and anistropic strains. (10.1117/12.3104760)
  DOI : 10.1117/12.3104760
• SCALMU: Synthetically-trained Coupling of Adaptive Learned Multiplicative Updates for Hyperspectral-Multispectral Fusion
  
  • Xu Xinxin
  • Gousseau Yann
  • Kervazo Christophe
  • Ladjal Saïd
  IEEE Transactions on Geoscience and Remote Sensing, Institute of Electrical and Electronics Engineers, 2026. HyperSpectral-MultiSpectral Image (HSI-MSI) fusion aims to recover a high-resolution hyperspectral image from a low-resolution HSI and a high-resolution MSI. Classical methods such as Coupled Nonnegative Matrix Factorization (CNMF) benefit from a strong physical interpretability but suffer from inferior results compared to their deep-learning counterparts. To address this limitation, we propose SCALMU (Synthetically-trained Coupling of Adaptive Learned Multiplicative Updates), a novel blind unrolled neural network architecture that integrates adaptive learnable matrices within the classical framework of CNMF multiplicative updates, improving its results. Due to its architectural proximity with CNMF, the resulting algorithm preserves physical interpretability and nonnegativity constraints. To overcome the scarcity of supervised training data, we generate a synthetic HSI-MSI dataset using the dead leaves model and train SCALMU end-to-end under synthetic supervision. Experiments on several datasets show that SCALMU outperforms state-of-the-art methods and highlights the potential of blind fusion trained with synthetic data. The code is available at \url{https://github.com/xinxinxu99/SCALMU.git} (10.1109/TGRS.2026.3712501)
  DOI : 10.1109/TGRS.2026.3712501
• Automatic Discovery of Disease Subgroups by Contrasting with Healthy Controls
  
  • Louiset Robin
  • Duchesnay Edouard
  • Dufumier Benoit
  • Grigis Antoine
  • Gori Pietro
  Data Mining and Knowledge Discovery, Springer, 2026.