Publications

2026

• Study of Gender Bias in Multimodal Human-Agent Collaboration: between observation and simulation
  
  • Elodie Etienne
  • Clavel Chloé
  • Ochs Magalie
  , 2026.
• Layer Collapse Can be Induced by Unstructured Pruning
  
  • Liao Zhu
  • Quétu Victor
  • Nguyen Van-Tam
  • Tartaglione Enzo
  Transactions on Machine Learning Research, [Amherst Massachusetts]: OpenReview.net, 2022, 2026. <div><p>Unstructured pruning is a popular compression method for efficiently reducing model parameters. However, while it effectively decreases the number of parameters, it is commonly believed that unstructured pruning cannot shorten the computational critical path, i.e., the maximum number of layers traversed during forward propagation.</p><p>In this paper, we study when and how unstructured pruning can yield structural effects. For rectifier-activated networks, we introduce the notion of neuron entropy, which quantifies the degree of nonlinearity utilization. We show that magnitude-based pruning naturally lowers this entropy, sometimes down to zero-entropy layers that become linearizable and can thus be removed. Building on this insight, we propose a method that leverages "unstructured" pruning to favor sparsity in low-entropy layers, enabling their complete removal. We validate the phenomenon across CNNs, Vision Transformers, and NLP models: unstructured pruning can induce effective layer removal with little or no performance degradation in over-parameterized networks. Our code is available at https://github.com/ZhuLIAO001/NEPENTHE.git.</p></div>
• Reconfigurable intelligent surfaces (RIS) using NOMA with energy harvesting from vibrations for 6G
  
  • Boujemaa Hatem
  • Alhussein Musaed
  • Rekaya Ghaya
  Annals of Telecommunications - annales des télécommunications, Springer, 2026, 81 (1-2), pp.121-131. Reconfigurable Intelligent Surfaces (RIS) enhance wireless communication by dynamically controlling electromagnetic waves. When combined with Non-Orthogonal Multiple Access (NOMA), RIS optimizes spectrum usage, allowing multiple users to share the same frequency via power domain multiplexing. This integration improves network capacity, spectral efficiency, and signal quality while reducing interference. Adding vibration-based energy harvesting to RIS with NOMA enables sustainable, autonomous operation in power-limited environments. This paper examines the potential of integrating RIS, NOMA, and vibration energy harvesting to advance sustainable, efficient wireless communication. (10.1007/s12243-025-01095-z)
  DOI : 10.1007/s12243-025-01095-z
• SIBH structure integration approaches and building blocks for next generation of transceivers on Indium Phosphide
  
  • Prado de la Cruz Shirley
  , 2026. Data traffic has significantly increased in recent years and is expected to keep rising due to bandwidth-intensive services such as Artificial Intelligence (AI) and high-definition gaming. One approach to enhance transmission data rates is to develop transceivers based on photonic integrated circuits (PICs), which are the key elements responsible for transmitting and receiving signals in Passive Optical Networks (PONs). The fundamental components forming a PIC, known as building blocks (BBs), are the focus of this work.Three approaches were investigated to improve the transmission capabilities of next-generation transceivers. The first concerns the development of SIBH technology used for the fabrication of active BBs. A one-step SIBH structure incorporating a blocking layer was developed. Several studies were conducted to evaluate the influence of growth temperature and layer thickness, with the optimal parameters identified as a growth temperature of 520 °C and a thickness of 40 nm. Additional investigations were also carried out on alternative materials for the semi-insulating layers of the SIBH structure, such as InP:Be and InP:Ru.The second approach focuses on passive BBs. Fabrication processes were developed to enable the monolithic integration of passive and active components despite their two different waveguide architectures (SIBH and deep-ridge). The main challenge encountered was the formation of polycrystals due to the SIBH regrowth, resulting in propagation losses exceeding 10 dB/cm for 1.5 µm deep-ridge waveguides. This issue was successfully resolved using an improved second fabrication process.Finally, a new InP (de)multiplexer structure based on a tilted angled multimode interferometer (AMMI) was developed. The device exhibits a compact footprint of 0,09 mmµ^2µ, a crosstalk level of 14.35 dB, and an insertion loss of 8 dB. (10.70675/57eec20fz9626z4acdza586z2ddef81a9013)
  DOI : 10.70675/57eec20fz9626z4acdza586z2ddef81a9013
• Random Stinespring superchannel: converting channel queries into dilation isometry queries
  
  • Girardi Filippo
  • Mele Francesco Anna
  • Zhao Haimeng
  • Fanizza Marco
  • Lami Ludovico
  , 2025. The recently introduced random purification channel, which converts $n$ copies of an arbitrary mixed quantum state into $n$ copies of the same uniformly random purification, has emerged as a powerful tool in quantum information theory. Motivated by this development, we introduce a channel-level analogue, which we call the random Stinespring superchannel. This consists in a procedure to transform $n$ parallel queries of an arbitrary quantum channel into $n$ parallel queries of the same uniformly random Stinespring isometry, via universal encoding and decoding operations that are efficiently implementable. When the channel is promised to have Choi rank at most $r$, the procedure can be tailored to yield a Stinespring environment of dimension $r$. As a consequence, quantum channel learning reduces to isometry learning, yielding a simple channel learning algorithm, based on existing isometry learning protocols, that matches the performance of the two recently proposed channel tomography algorithms. Complementarily, whereas the optimality of these algorithms had previously been established only up to a logarithmic factor in the dimension, we close this gap by removing this logarithmic factor from the lower bound. Taken together, our results fully establish the optimality of these recently introduced channel learning algorithms, showing that the optimal query complexity of learning a quantum channel with input dimension $d_A$, output dimension $d_B$, and Choi rank $r$ is $Θ(d_A d_B r)$.
• Non-iid hypothesis testing: from classical to quantum
  
  • de Palma Giacomo
  • Fanizza Marco
  • Mowry Connor
  • O'Donnell Ryan
  , 2025. We study hypothesis testing (aka state certification) in the non-identically distributed setting. A recent work (Garg et al. 2023) considered the classical case, in which one is given (independent) samples from $T$ unknown probability distributions $p_1, \dots, p_T$ on $[d] = \{1, 2, \dots, d\}$, and one wishes to accept/reject the hypothesis that their average $p_{\mathrm{avg}}$ equals a known hypothesis distribution $q$. Garg et al. showed that if one has just $c = 2$ samples from each $p_i$, and provided $T \gg \frac{\sqrt{d}}{ε^2} + \frac{1}{ε^4}$, one can (whp) distinguish $p_{\mathrm{avg}} = q$ from $d_{\mathrm{TV}}(p_{\mathrm{avg}},q) &gt; ε$. This nearly matches the optimal result for the classical iid setting (namely, $T \gg \frac{\sqrt{d}}{ε^2}$). Besides optimally improving this result (and generalizing to tolerant testing with more stringent distance measures), we study the analogous problem of hypothesis testing for non-identical quantum states. Here we uncover an unexpected phenomenon: for any $d$-dimensional hypothesis state $σ$, and given just a single copy ($c = 1$) of each state $ρ_1, \dots, ρ_T$, one can distinguish $ρ_{\mathrm{avg}} = σ$ from $D_{\mathrm{tr}}(ρ_{\mathrm{avg}},σ) &gt; ε$ provided $T \gg d/ε^2$. (Again, we generalize to tolerant testing with more stringent distance measures.) This matches the optimal result for the iid case, which is surprising because doing this with $c = 1$ is provably impossible in the classical case. We also show that the analogous phenomenon happens for the non-iid extension of identity testing between unknown states. A technical tool we introduce may be of independent interest: an Efron-Stein inequality, and more generally an Efron-Stein decomposition, in the quantum setting. (10.48550/arXiv.2510.06147)
  DOI : 10.48550/arXiv.2510.06147
• Random purification channel for passive Gaussian bosons
  
  • Mele Francesco Anna
  • Girardi Filippo
  • Chen Senrui
  • Fanizza Marco
  • Lami Ludovico
  , 2025. (10.48550/arXiv.2512.16878)
  DOI : 10.48550/arXiv.2512.16878
• Efficient learning of bosonic Gaussian unitaries
  
  • Fanizza Marco
  • Iyer Vishnu
  • Lee Junseo
  • Mele Antonio A.
  • Mele Francesco A.
  , 2025. Bosonic Gaussian unitaries are fundamental building blocks of central continuous-variable quantum technologies such as quantum-optic interferometry and bosonic error-correction schemes. In this work, we present the first time-efficient algorithm for learning bosonic Gaussian unitaries with a rigorous analysis. Our algorithm produces an estimate of the unknown unitary that is accurate to small worst-case error, measured by the physically motivated energy-constrained diamond distance. Its runtime and query complexity scale polynomially with the number of modes, the inverse target accuracy, and natural energy parameters quantifying the allowed input energy and the unitary's output-energy growth.<p>The protocol uses only experimentally friendly photonic resources-coherent and squeezed probes, passive linear optics, and heterodyne/homodyne detection. We then employ an efficient classical post-processing routine that leverages a symplectic regularization step to project matrix estimates onto the symplectic group. In the limit of unbounded input energy, our procedure attains arbitrarily high precision using only 2m + 2 queries, where m is the number of modes. To our knowledge, this is the first provably efficient learning algorithm for a multiparameter family of continuous-variable unitaries.</p> (10.48550/arXiv.2510.05531)
  DOI : 10.48550/arXiv.2510.05531
• Unrolled Multiplicative Updates for Nonnegative Matrix Factorization applied to Hyperspectral Unmixing
  
  • Kervazo Christophe
  • Cohen Jérémy E.
  , 2026. HyperSpectral Unmixing (HSU), the problem of separating mixed spectra of overlapping materials in a hyperspectral image, has motivated dedicated algorithmic developments in the last two decades. On the one hand, traditional model-based algorithms frequently guarantee interpretable results. On the other hand, deep-learning-based approaches are often faster at inference time and may obtain better empirical results. This work utilizes the strengths of both approaches by building on the deep unrolling paradigm. Our contribution is twofold. First, we propose two new algorithms based on deep unrolling of the well-known Multiplicative Updates. The first, coined Non-Adaptive Learned Multiplicative Updates (NALMU), adopts a simple element-wise multiplicative scheme. The second, called Recursive Adaptive Learned Multiplicative Updates (RALMU), has more flexible updates and better take into account the spatial correlations in the abundances. Second, we relate NALMU to the minimization of an explicit cost function under some assumptions. Such guarantees are unique in the HSU field. NALMU and RALMU are tested on astrophysics and remote sensing datasets. They outperform the other deep learning-based HSU algorithms and classical iterative schemes for the endmember estimates and obtain competitive results for the abundance estimates, even when trained in a self-supervised way. The code used in this paper will be made available upon publication.
• The NPA hierarchy does not always attain the commuting operator value
  
  • Fanizza Marco
  • Kroell Larissa
  • Mehta Arthur
  • Paddock Connor
  • Rochette Denis
  • Slofstra William
  • Zhao Yuming
  , 2025. We show that it is undecidable to determine whether the commuting operator value of a nonlocal game is strictly greater than 1/2. Specifically, there is a computable mapping from Turing machines to /boolean constraint system (BCS) nonlocal games in which the halting property of the machine is encoded as a decision problem for the commuting operator value of the game. As a corollary, there is a BCS game for which the value of the Navascués-Pironio-Acín (NPA) hierarchy does not attain the commuting operator value at any finite level. (10.48550/arXiv.2510.04943)
  DOI : 10.48550/arXiv.2510.04943
• Receiver Noise Calibration in CV-QKD accounting for Noise Dynamics
  
  • Ricard Guillaume
  • Jaouën Yves
  • Alléaume Romain
  , 2025, pp.043287. Continuous-Variable Quantum Key Distribution (CV-QKD) relies on accurate noise calibration at the receiver to ensure the security of quantum communication. Traditional calibration methods often oversimplify noise characteristics, neglecting the impact of local oscillator (LO) noise and the critical role of noise spectral properties, which can lead to imprecise Shot Noise Calibration (SNC). Our contributions are threefold: 1) we propose an operational framework for calibration, relying on the notion of stationarity 2) in this framework, we give a method allowing us to derive the optimal calibration duration for a given experiment 3) leveraging our knowledge of noise spectral properties, we introduce a novel SNC method. This work also formalizes the calibration procedures, addressing implicit assumptions and providing a better foundation for the certification of CV-QKD protocols, of which calibration is a fundamental part. We demonstrate that our improved calibration technique offers higher performance and higher tolerance to receiver imperfections, which can enhance the performance and cost-effectiveness of CV-QKD systems. (10.48550/arXiv.2509.07549)
  DOI : 10.48550/arXiv.2509.07549
• Computational aspects of the trace norm contraction coefficient
  
  • Delsol Idris
  • Fawzi Omar
  • Kochanowski Jan
  • Ramachandran Akshay
  , 2026. We show that approximating the trace norm contraction coefficient of a quantum channel within a constant factor is NP-hard. Equivalently, this shows that determining the optimal success probability for encoding a bit in a quantum system undergoing noise is NP-hard. This contrasts with the classical analogue of this problem that can clearly be solved efficiently. We also establish the NP-hardness of deciding if the contraction coefficient is equal to 1, i.e., the channel can perfectly preserve a bit. As a consequence, deciding if a non-commutative graph has an independence number of at least 2 is NPhard. In addition, we establish a converging hierarchy of semidefinite programming upper bounds on the contraction coefficient.
• Phy2-ExposNet: A Physics-Informed Neural Network for Urban EMF Exposure Mapping
  
  • Li Shuangning
  • Zhang Yarui
  • Wang Shanshan
  • Wiart Joe
  IEEE Open Journal of Antennas and Propagation, IEEE, 2026, pp.1-13. Accurate electromagnetic field (EMF) exposure mapping is critical for wireless network planning, environmental monitoring, and the deployment of next generation communication systems. The mapping results can be converted into the form of a radio map, a key technology in digital twin communication systems, which describes the wireless signal propagation characteristics at every location in a specific area. Existing deep learning approaches treat propagation estimation as a pure regression problem and do not enforce physical consistency in the predicted fields. In this paper, we propose Phy2-ExposNet, a novel neural network framework that decouples exposure mapping into a physics-informed estimation stage and a transformer-based residual refinement stage. It first estimates the fields under two physical constraints and then refines the resulting exposure map by capturing long-range interactions and complex spatial propagation patterns. Experiments demonstrate that the proposed method achieves lower estimation error while significantly reducing model complexity compared to existing approaches. It achieves around 15% relative error reduction over baselines, while using over 80% fewer parameters than conventional physics-informed models. Ablation results further reveal that the physics-informed design is crucial for capturing complex propagation effects, particularly in boundary and shadow regions. (10.1109/OJAP.2026.3707005)
  DOI : 10.1109/OJAP.2026.3707005
• Simulation-Driven Railway Delay Prediction: An Imitation Learning Approach
  
  • Elliker Clément
  • Read Jesse
  • Vanier Sonia
  • Bifet Albert
  , 2025, 40 (25), pp.20977-20984. Reliable prediction of train delays is essential for enhancing the robustness and efficiency of railway transportation systems. In this work, we reframe delay forecasting as a stochastic simulation task, modeling state-transition dynamics through imitation learning. We introduce Drift-Corrected Imitation Learning (DCIL), a novel self-supervised algorithm that extends DAgger by incorporating distance-based drift correction, thereby mitigating covariate shift during rollouts without requiring access to an external oracle or adversarial schemes. Our approach synthesizes the dynamical fidelity of event-driven models with the representational capacity of data-driven methods, enabling uncertainty-aware forecasting via Monte Carlo simulation. We evaluate DCIL using a comprehensive real-world dataset from \textsc{Infrabel}, the Belgian railway infrastructure manager, which encompasses over three million train movements. Our results, focused on predictions up to 30 minutes ahead, demonstrate superior predictive performance of DCIL over traditional regression models and behavioral cloning on deep learning architectures, highlighting its effectiveness in capturing the sequential and uncertain nature of delay propagation in large-scale networks. (10.1609/aaai.v40i25.39239)
  DOI : 10.1609/aaai.v40i25.39239
• I-INR: Iterative Implicit Neural Representations
  
  • Haider Ali
  • Ali Muhammad Salman
  • Qamar Maryam
  • Khalil Tahir
  • Kim Soo Ye
  • Oh Jihyong
  • Tartaglione Enzo
  • Bae Sung-Ho
  , 2026, 40 (6), pp.503, 4520-4528. Implicit Neural Representations (INRs) have revolutionized signal processing and computer vision by modeling signals as continuous, differentiable functions parameterized by neural networks. However, INRs are prone to the spectral bias problem, limiting their ability to retain high-frequency information, and often struggle with noise robustness. Motivated by recent trends in iterative refinement processes, we propose Iterative Implicit Neural Representations (I-INRs). This novel plug-and-play framework iteratively refines signal reconstructions to restore high-frequency details, improve noise robustness, and enhance generalization, ultimately delivering superior reconstruction quality. I-INRs integrate seamlessly into existing INR architectures with only a 0.5–2% increase in parameters. During reconstruction, the iterative refinement adds just 0.8–1.6% additional FLOPs over the baseline while delivering a substantial performance boost of up to +2.0 PSNR. Extensive experiments demonstrate that I-INRs consistently outperform WIRE, SIREN, and Gauss across various computer vision tasks, including image fitting, image denoising, and object occupancy prediction. (10.1609/aaai.v40i6.4245)
  DOI : 10.1609/aaai.v40i6.4245
• Convergence of the Cumulant Expansion and Polynomial-Time Algorithm for Weakly Interacting Fermions
  
  • Chen Hongrui
  • Rouzé Cambyse
  • Chen Jielun
  • Jiang Jiaqing
  • Scalet Samuel
  • Zhan Yongtao
  • Chan Garnet Kin-Lic
  • Ying Lexing
  • Tong Yu
  , 2025. We propose a randomized algorithm to compute the log-partition function of weakly interacting fermions with polynomial runtime in both the system size and precision. Although weakly interacting fermionic systems are considered tractable for many computational methods such as the diagrammatic quantum Monte Carlo, a mathematically rigorous proof of polynomial runtime has been lacking. In this work we first extend the proof techniques developed in previous works for proving the convergence of the cumulant expansion in periodic systems to the non-periodic case. A key equation used to analyze the sum of connected Feynman diagrams, which we call the tree-determinant expansion, reveals an underlying tree structure in the summation. This enables us to design a new randomized algorithm to compute the log-partition function through importance sampling augmented by belief propagation. This approach differs from the traditional method based on Markov chain Monte Carlo, whose efficiency is hard to guarantee, and enables us to obtain a algorithm with provable polynomial runtime. (10.48550/arXiv.2512.12010)
  DOI : 10.48550/arXiv.2512.12010
• Free space optical communications in fog: comparing wavelengths for intersymbol interference resilience
  
  • Breton Alberto
  • Sorrente Béatrice
  • Fade Julien
  • Silva Anabela Da
  • Grillot Frédéric
  , 2026, 13890, pp.138900V. Free-space optical (FSO) communications at 1.55 μm are gaining increasing recognition due to their high data rates and smaller beam divergence compared to radio-frequency systems. However, FSO links are highly sensitive to adverse atmospheric conditions and turbulence, which limits their practicality in urban environments, particularly in the presence of fog. To mitigate these impairments, the use of longer wavelengths has been proposed. The aim of this work is to investigate the effects of fog on optical communication links operating at 1.55 μm and 10.3 μm. The study focuses on the temporal spreading of transmitted optical pulses caused by multiple scattering interactions between photons and water droplets. This pulse broadening can significantly limit the achievable data rate due to inter-symbol interference (ISI). A radiative transfer model based on the radiative transfer equation (RTE) is employed to compute the impulse response of a foggy atmospheric slab. These impulse responses are then applied to an amplitude-modulated optical signal to assess system performance under fog conditions. The results show that the link operating at 10.2 μm is less affected by temporal spreading than the 1.55 μm link, demonstrating improved robustness against fog-induced ISI degradation. (10.1117/12.3079603)
  DOI : 10.1117/12.3079603
• Scalable Information Theoretic Evaluation of the Rank Statistics in Side-Channel Attacks
  
  • Béguinot Julien
  • Rioul Olivier
  • Masure Loïc
  • Standaert François-Xavier
  • Cheng Wei
  • Guilley Sylvain
  IACR Transactions on Cryptographic Hardware and Embedded Systems, IACR, 2026, 2026 (1), pp.53-81. Evaluating the security of a device against side-channel attacks is a difficult task. One prominent strategy for this purpose is to characterize the distribution of the rank of the correct key among the different key hypotheses produced by a maximum likelihood attack, depending on the number of measured traces. In practice, evaluators can estimate some statistics of the rank that are used as security indicators—e.g., the arithmetic and geometric mean rank, the median rank, the α-marginal guesswork, or the success rate of level L. Yet, a direct estimation becomes time-consuming as security levels increase.In this work, we provide new bounds on these figures of merit in terms of the mutual information between the secret and its side-channel leakages. These bounds provide theoretical insights on the evolution of the figures of merit in terms of noise level, computational complexity (how many keys are evaluated) and data complexity (how many side-channel traces are used for the attack). To the best of our knowledge, these bounds are the first to formally characterize security guarantees that depend on the computational power of the adversary, based on a measure of their informational leakages. It follows that our results enable fast shortcut formulas for the certification laboratories, potentially enabling them to speed up the security evaluation process. We demonstrate the tightness of our bounds on both synthetic traces (in a controlled environment) and real-world traces from two popular datasets (Aisylab/AES_HD and SMAesH). (10.46586/tches.v2026.i1.53-81)
  DOI : 10.46586/tches.v2026.i1.53-81
• Docker does not Guarantee Reproducibility
  
  • Malka Julien
  • Zacchiroli Stefano
  • Zimmermann Théo
  , 2026. <div><p>The reproducibility of software environments is a critical concern in modern software engineering, with ramifications ranging from the effectiveness of collaboration workflows to software supply chain security and scientific reproducibility. Containerization technologies like Docker address this problem by encapsulating software environments into shareable filesystem snapshots known as images. While Docker is frequently cited in the literature as a tool that enables reproducibility in theory, the extent of its guarantees and limitations in practice remains under-explored.</p><p>In this work, we address this gap through two complementary approaches. First, we conduct a systematic literature review to examine how Docker is framed in scientific discourse on reproducibility and to identify documented best practices for writing Dockerfiles enabling reproducible image building. Then, we perform a large-scale empirical study of 5298 Docker builds collected from GitHub workflows. By rebuilding these images and comparing the results with their historical counterparts, we assess the real reproducibility of Docker images and evaluate the effectiveness of the best practices identified in the literature.</p></div>
• Thinking Before Constraining: A Unified Decoding Framework for Large Language Models
  
  • Nguyen Ngoc Trinh Hung
  • Silva Alonso
  • Zumot Laith
  • Tupikina Liubov
  • Aghasaryan Armen
  • Alam Mehwish
  , 2026. Natural generation allows Language Models (LMs) to produce free-form responses with rich reasoning, but the lack of guaranteed structure makes outputs difficult to parse or verify. Structured generation, or constrained decoding, addresses this drawback by producing content in standardized formats such as JSON, ensuring consistency and guaranteed-parsable outputs, but it can inadvertently restrict the model's reasoning capabilities. In this work, we propose a simple approach that combines the advantages of both natural and structured generation. By allowing LLMs to reason freely until specific trigger tokens are generated, and then switching to structured generation, our method preserves the expressive power of natural language reasoning while ensuring the reliability of structured outputs. We further evaluate our approach on several datasets, covering both classification and reasoning tasks, to demonstrate its effectiveness, achieving a substantial gain of up to 27% in accuracy compared to natural generation, while requiring only a small overhead of 10-20 extra tokens.
• Towards Robust Secure Compilation in Presence of Speculative Execution
  
  • Clément Léopold
  • Kühne Ulrich
  • Brandner Florian
  • Pacalet Renaud
  , 2026. Time-based side-channel attacks have been known since 1996. Many such attacks use the differences in execution time between program executions caused e.g. by cache hits/misses or non-constant-time instructions. To counter them, programmers use constant time programming, a set of empirical rules that are supposed to avoid a secret leaking via execution time variations. However, until the discovery of the Spectre attack, the verifications were based on the instruction set architecture (ISA) specification that does not take into account speculative execution. Speculation widely varies between actual ISA implementations. Creating a precise model is mostly impractical for verification. In our current work, we incorporate speculative semantics into the compiler CompCert. An additional pass has been added that inserts speculation barriers in order to counter information leakage from transient executions. We have succeeded to prove the preservation of constant time for a simple and overly aggressive strategy, which inserts such barriers after each branch instruction. In this paper, we briefly describe the challenges that remain to be addressed.
• LELA: an LLM-based Entity Linking Approach with Zero-Shot Domain Adaptation
  
  • Haffoudhi Samy
  • Suchanek Fabian M
  • Holzenberger Nils
  , 2026. Entity linking (mapping ambiguous mentions in text to entities in a knowledge base) is a foundational step in tasks such as knowledge graph construction, question-answering, and information extraction. Our method, LELA, is a modular coarse-to-fine approach that leverages the capabilities of large language models (LLMs), and works with different target domains, knowledge bases and LLMs, without any fine-tuning phase. Our experiments across various entity linking settings show that LELA is highly competitive with fine-tuned approaches, and substantially outperforms the non-fine-tuned ones.
• Comic character recognition using graph embedding in graph representations
  
  • Kajla Nadeem Iqbal
  • Saad Missen Malik Muhammad
  • Pasha Maruf
  • Coustaty Mickael
  • Luqman Muhammad Muzzamil
  • Belbachir Faiza
  • Mushtaq Maria
  International Journal on Document Analysis and Recognition, Springer Verlag, 2026, 29 (2), pp.483-493. (10.1007/s10032-025-00568-0)
  DOI : 10.1007/s10032-025-00568-0
• 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