Publications

2025

• A gem5-Based Framework for RISC-V Security Analysis
  
  • Khan Mahreen
  • Mushtaq Maria
  • Pacalet Renaud
  • Apvrille Ludovic
  , 2025. <div><p>1] cheng2024,Evict+Spec+Time: Exploiting Out-of-Order Execution to Improve Cache Attacks Enhanced variant of Evict+Time attack combining eviction, speculation, and timing originally tested for x86. We tested it on RISC-V architecture.</p></div>
• Annealed Winner-Takes-All for Motion Forecasting
  
  • Xu Yihong
  • Letzelter Victor
  • Chen Mickaël
  • Zablocki Éloi
  • Cord Matthieu
  , 2025. In autonomous driving, motion prediction aims at forecasting the future trajectories of nearby agents, helping the ego vehicle to anticipate behaviors and drive safely. A key challenge is generating a diverse set of future predictions, commonly addressed using data-driven models with Multiple Choice Learning (MCL) architectures and Winner-Takes-All (WTA) training objectives. However, these methods face initialization sensitivity and training instabilities. Additionally, to compensate for limited performance, some approaches rely on training with a large set of hypotheses, requiring a post-selection step during inference to significantly reduce the number of predictions. To tackle these issues, we take inspiration from annealed MCL, a recently introduced technique that improves the convergence properties of MCL methods through an annealed Winner-Takes-All loss (aWTA). In this paper, we demonstrate how the aWTA loss can be integrated with state-of-the-art motion forecasting models to enhance their performance using only a minimal set of hypotheses, eliminating the need for the cumbersome post-selection step. Our approach can be easily incorporated into any trajectory prediction model normally trained using WTA and yields significant improvements. To facilitate the application of our approach to future motion forecasting models, the code is made publicly available: https://github.com/valeoai/MF_aWTA.
• From trustworthy AI to technical standards - The distinctive European approach to artificial intelligence regulation
  
  • Gornet Mélanie
  , 2025. Europe has been at the forefront of Artificial Intelligence (AI) ethics, developing non-binding charters and principles on "trustworthy'' AI. The term "trustworthiness'' is used by Europe to designate AI systems that are "ethical'', "legal'' and "technically robust''. Europe has supplemented these non-binding principles with a binding regulation on AI, known as the AI Act. The AI Act is one of the world's first comprehensive frameworks for regulating AI systems across different industries and use cases, focusing on safety and protection of fundamental rights. The AI Act relies, for operational questions, mostly on technical standards that are in the course of development. The European approach thus combines three layers of regulatory instruments: AI ethics charters, the AI Act and technical standards.The standardisation approach is traditional in product safety, but under the AI Act, standards are also expected to address fundamental rights concerns. To avoid making hard normative choices, standardisation organisations are playing it safe, developing standards which remain at a high-level. Moreover, under the AI Act, the responsibility for developing technical standards is delegated to private standardisation bodies, where large multinational companies are over-represented and hold significant influence. These standards are also often locked behind paywalls, although the situation may evolve in the coming years after a recent case law from the Court of Justice of the European Union. Standardisation experts therefore face pressures to deliver standards on time and of good quality. (10.70675/0cb526b5z9f0cz491ezbf98z1427db363cbb)
  DOI : 10.70675/0cb526b5z9f0cz491ezbf98z1427db363cbb
• Equivariant Denoisers for Image Restoration
  
  • Renaud Marien
  • Leclaire Arthur
  • Papadakis Nicolas
  , 2025, pp.227 - 240. One key ingredient of image restoration is to define a realistic prior on clean images to complete the missing information in the observation. State-of-the-art restoration methods rely on a neural network to encode this prior. Moreover, typical image distributions are invariant to some set of transformations, such as rotations or flips. However, most deep architectures are not designed to represent an invariant image distribution. Recent works have proposed to overcome this difficulty by including equivariance properties within a Plug-and-Play paradigm. In this work, we propose a unified framework named Equivariant Regularization by Denoising (ERED) based on equivariant denoisers and stochastic optimization. We analyze the convergence of this algorithm and discuss its practical benefit. (10.1007/978-3-031-92366-1_18)
  DOI : 10.1007/978-3-031-92366-1_18
• Generation of frequency entanglement with an effective quantum dot-waveguide two-photon quadratic interaction
  
  • Meguebel Mohamed
  • Federico Maxime
  • Felicetti Simone
  • Belabas Nadia
  • Fabre Nicolas
  , 2025. Light–matter interactions with quantum dots have been extensively studied to harness key quantum properties of photons, such as indistinguishability and entanglement. In this theoretical work, we exploit the atomic-like four-level structure of a quantum dot coupled to a waveguide to model a shaping frequency entangling gate (ShaFrEnGa) for single photons. Our approach is based on the identification of input frequencies and an atomic level structure for which frequency-dependent one-photon transitions are adiabatically eliminated, while frequency-dependent two-photon transitions are resonantly enhanced. The frequency entanglement performance of the gate are analyzed using a Schmidt decomposition for continuous variables, revealing a trade-off between entanglement generation efficiency and entanglement quality. We further demonstrate the use of the ShaFrEnGa for the generation of entangled frequency qudit states.
• Computer vision-based foot contact detection for long jump using a monocular normal-speed camera
  
  • Fang Yangtao
  • Gan Qi
  • Nguyen Sao Mai
  , 2025.
• Correct-by-construction requirement decomposition
  
  • Sun Minghui
  • Bakirtzis Georgios
  • Jafarzadeh Hassan
  • Fleming Cody
  Software and Systems Modeling, Springer Verlag, 2025, pp.1-16. In systems engineering, accurately decomposing requirements is crucial for creating well-defined and manageable system components, particularly in safety-critical domains. Despite the critical need, rigorous, top-down methodologies for effectively breaking down complex requirements into precise, actionable sub-requirements are scarce, especially compared to the wealth of bottom-up verification techniques. Addressing this gap, we introduce a formal decomposition for contract-based design that guarantees the correctness of decomposed requirements if specific conditions are met. Our (semi-)automated methodology augments contract-based design with reachability analysis and constraint programming to systematically identify, verify, and validate sub-requirements representable by continuous bounded sets---continuous relations between real-valued inputs and outputs. We demonstrate the efficacy and practicality of a correct-by-construction approach through a comprehensive case study on a cruise control system, highlighting how our methodology improves the interpretability, tractability, and verifiability of system requirements. (10.1007/s10270-025-01291-4)
  DOI : 10.1007/s10270-025-01291-4
• Interactive Sketch-based Modeling of Braided Hair
  
  • Jetti Hari Hara Gowtham
  • Parakkat Amal Dev
  , 2025, pp.1-2. Hair braids are widely used in various games and animated movies, thanks to their simplified representation and ease of animation. However, the existing research on modeling braids often relies on a limited dictionary of commonly seen hair braid patterns, constraining artists' ability to experiment by creating imaginary or creative hair braids. In this paper, we introduce a simple sketch-based interface for creating arbitrary hair braids. Our method employs a two-stage framework that first interprets a user-drawn sketch to extract the braid pattern. To accommodate arbitrarily drawn sketches, we then use a physics-inspired simulation to generate visually pleasing braids. In addition to automatically generating braids, our system allows users to interactively refine the braid pattern to create braids that match the user's imagination, facilitating experimentation and exploration of different braid structures. (10.2312/egp.20251027)
  DOI : 10.2312/egp.20251027
• Multi-client Functional Encryption with Public Inputs and Strong Security
  
  • Nguyen Ky
  • Phan Duong Hieu
  • Pointcheval David
  , 2025, 15676, pp.68-101. Recent years have witnessed a significant development for functional encryption (FE) in the multi-user setting, particularly with multi-client functional encryption (MCFE). The challenge becomes more important when combined with access control, such as attribute-based encryption (ABE), which was actually not covered syntactically by the public-key FE nor semantically by the secret-key MCFE frameworks. On the other hand, as for complex primitives, many works have studied the admissibility of adversaries to ensure that the security model encompasses all real threats of attacks. 1. At a conceptual level, by adding a public input to FE/MCFE, we cover many previous primitives, notably attribute-based function classes. Furthermore, with the strongest admissibility for inner-product functionality, our framework is quite versatile, as it encrypts multiple sub-vectors, allows repetitions and corruptions, and eventually also encompasses public-key FE and classical ABE, bridging the private setting of MCFE with the public setting of FE and ABE. 2. Finally, we propose an MCFE with public inputs with the class of functions that combines inner-products (on private inputs) and attribute-based access-control (on public inputs) for LSSS policies. We achieve the first AB-MCFE for inner products with strong admissibility (from Nguyen et al., ACNS’23) and with adaptive security. In the end, our concrete MCFE leads to MIFE for inner products, public-key single-input inner- product FE with LSSS key-policy, and KP-ABE for LSSS, with adaptive security. Previous AB-MCFE constructions are either restricted in terms of weaker admissibility (Nguyen et al., ASIACRYPT’22) or considers a slightly larger functionality of attribute-weighted sum but with only selective security (Agrawal et al., CRYPTO’23). (10.1007/978-3-031-91826-1_3)
  DOI : 10.1007/978-3-031-91826-1_3
• Securing Cooperative Vehicular Platooning with a Set of Reinforced Checks
  
  • Braiteh Farah-Emma
  • Bassi Francesca
  • Khatoun Rida
  , 2025. Platooning enhances road safety and alleviates traffic congestion by enabling vehicles to travel closely together and maneuver in a coordinated manner. This coordination is facilitated by vehicle-to-vehicle (V2V) communications, which, unfortunately, also expose the platoon to potential cyberattack risks. In this paper, we present a novel platoon joining protocol, with a particular emphasis on the enrollment phase. We demonstrate that an attacker can disrupt the platoon’s formation or stability by falsely joining, without actually maneuvering into the platoon. To mitigate this risk, we propose robust physical challenges and data-consistency countermeasures that reinforce both the stability and integrity of the platoon. Simulations using Plexe validate the security of the designed protocol, as verified and confirmed thorough security checks. (10.1109/IWCMC65282.2025.11059583)
  DOI : 10.1109/IWCMC65282.2025.11059583
• Infusion: Internal Diffusion for Inpainting of Dynamic Textures and Complex Motion
  
  • Cherel Nicolas
  • Almansa Andrés
  • Gousseau Yann
  • Newson Alasdair
  , 2023, pp.446-450. Video inpainting is the task of filling a desired region in a video in a visually convincing manner. It is a very challenging task due to the high dimensionality of the signal and the temporal consistency required for obtaining convincing results. Recently, diffusion models have shown impressive results in modeling complex data distributions, including images and videos. Diffusion models remain nonetheless very expensive to train and perform inference with, which strongly restrict their application to video. We show that in the case of video inpainting, thanks to the highly auto-similar nature of videos, the training of a diffusion model can be restricted to the video to inpaint and still produce very satisfying results. This leads us to adopt an internal learning approch, which also allows for a greatly reduced network size. We call our approach "Infusion": an internal learning algorithm for video inpainting through diffusion. Due to our frugal network, we are able to propose the first video inpainting approach based purely on diffusion. Other methods require supporting elements such as optical flow estimation, which limits their performance in the case of dynamic textures for example. We introduce a new method for efficient training and inference of diffusion models in the context of internal learning. We split the diffusion process into different learning intervals which greatly simplifies the learning steps. We show qualititative and quantitative results, demonstrating that our method reaches state-of-the-art performance, in particular in the case of dynamic backgrounds and textures. (10.1111/cgf.70070)
  DOI : 10.1111/cgf.70070
• Arithmetisation of the Floor Function and Its Applications to Homomorphic Cryptography
  
  • Berthet Pierre-Augustin
  • Tavernier Cédric
  , 2025. Cryptography has historically been based on integer arithmetic. Thus, there was no need to investigate functions related to real numbers or analysis, such as the fractional part or the floor function. The floor function has several applications in modern cryptography. Its arithmetisation can allow for the application of generic side-channel countermeasures, like masking, without being limited by the chosen representation of rationnal or real numbers. It has also some applications in Fully Homomorphic Encryption (FHE), either directly in CKKS, or indirectly, as an arithmetised floor function can be computed with FHE. A consequence is the possibility of protecting normalisation or discretisation operations in Machine Learning or Deep Learning. In this work, we perform the arithmetisation by adapting a Fourier series and speeding up its convergence by composing partial series with themselves.
• PerceptualLift: Using hatches to infer a 3D organic shape from a sketch
  
  • Butler Tara
  • Guehl Pascal
  • Parakkat Amal Dev
  • Cani Marie-Paule
  , 2025. In this work, we investigate whether artistic hatching, popular in pen-and-ink sketches, can be consistently perceived as a depth cue. We illustrate our results by presenting PerceptualLift, a modeling system that exploits hatching to create curved 3D shapes from a single sketch. We first describe a perceptual user study conducted across a diverse group of participants, which confirms the relevance of hatches as consistent clues for inferring curvature in the depth direction from a sketch. It enables us to extract geometrical rules that link 2D hatch characteristics, such as their direction, frequency, and magnitude, to the changes of depth in the depicted 3D shape. Built on these rules, we introduce PerceptualLift, a flexible tool to model 3D organic shapes by simply hatching over 2D hand-drawn contour sketches. (10.2312/exw.20251055)
  DOI : 10.2312/exw.20251055
• Memory attacks in network nonlocality and self-testing
  
  • Weilenmann Mirjam
  • Budroni Costantino
  • Navascués Miguel
  Quantum, Verein zur Förderung des Open Access Publizierens in den Quantenwissenschaften, 2025, 9, pp.1735. We study what can or cannot be certified in communication scenarios where the assumption of independence and identical distribution (iid) between experimental rounds fails. In this respect, we prove that membership tests for non-convex sets of correlations cannot be formulated in the non-iid regime. Similarly, it is impossible to self-test non-extreme quantum operations, such as mixed states, or noisy quantum measurements, unless one allows more than a single use thereof within the same experimental round. One consequence of our results is that non-classicality in causal networks without inputs cannot be experimentally demonstrated. By analyzing optimal non-iid strategies in the triangle scenario, we raise the need to take into account the prior communication required to set up a causal network. (10.22331/q-2025-05-06-1735)
  DOI : 10.22331/q-2025-05-06-1735
• Digital Persuasion: Understanding the Impact of Online Influencers on Public Opinion
  
  • Berjawi Omran
  • Khatoun Rida
  • Fenza Giuseppe
  , 2025.
• Anamorphism Beyond One-to-One Messaging: Public-Key with Anamorphic Broadcast Mode
  
  • Do Xuan Thanh
  • Persiano Giuseppe
  • Phan Duong Hieu
  • Yung Moti
  , 2025, 15603, pp.429-455. To date, Anamorphic Cryptography [EC22] has been developed to support adding an anamorphic message within a ciphertext carrying a primary message. The anamorphic message remains hidden even in the presence of a strong adversary that possesses the receiver’s key and/or determined the sent primary message. In this paper, we expand one-to-one encrypted anamorphic communication to one-to-many anamorphism, naturally assuming communication over a broadcast channel. What we show is that using a previously designed public-key encryption scheme, two things can happen: First, the receiver of an added hidden message may be a party different from the actual receiver (i.e., a shadow party) who has initially collaborated with the sender. Secondly, and perhaps more surprisingly, the receiving party need not be a singleton, and can be a number of different shadow (i.e., anonymous) groups, each receiving a different anamorphic message, where all these messages are extracted from a single one-receiver ciphertext. The idea of having multiple hidden channels to different shadow groups is highly handy if, for example, the anamorphic messages are warnings with operational instructions, sent to the groups and will be received by a group even if the adversary is able to temporarily cut off all but one members of a channel. More specifically, First, we motivate and formalize the notion of Public-Key Encryption with an Anamorphic Broadcast Mode. We then present, as an initial result of an independent interest, the first lattice-based construction of Anonymous Multi-Channel Broadcast Encryption. It is important to note here that all Multi-Channel Broadcast schemes to date are in the pairing-based setting (and are, thus, insecure against quantum adversaries). Finally, we show how to transform a strong form of anonymity (where the ciphertext also hides the number of channels) into a system with anamorphism in the multi-channel broadcast setting for the well-known Dual Regev Public-Key Encryption scheme. Specifically, we show that, given the public key for the Dual Regev encryption scheme, and a sequence of messages for the channels of broadcast scheme, it is possible to create a ciphertext that will carry the messages and is also a legitimate ciphertext for PK. (10.1007/978-3-031-91131-6_15)
  DOI : 10.1007/978-3-031-91131-6_15
• Associations of pulmonary microvascular blood volume with per cent emphysema and CT emphysema subtypes in the community: the MESA Lung study
  
  • Hermann Emilia
  • Motahari Amin
  • Hoffman Eric
  • Sun Yifei
  • Allen Norrina
  • Angelini Elsa
  • Bertoni Alain
  • Bluemke David
  • Gerard Sarah
  • Guo Junfeng
  • Kaczka David
  • Laine Andrew
  • Michos Erin
  • Nagpal Prashant
  • Pankow James
  • Smith Benjamin
  • Sack Coralynn
  • Stukovsky Karen Hinckley
  • Watson Karol
  • Wysoczanski Artur
  • Barr R Graham
  Thorax, BMJ Publishing Group, 2025, 80 (5), pp.309-317. Background Pulmonary microvasculature alterations are implicated in emphysema pathogenesis, but the association between pulmonary microvascular blood volume (PMBV) and emphysema has not been directly assessed at scale, and prior studies have used non-specific measures of emphysema. Methods The Multi-Ethnic Study of Atherosclerosis Lung Study invited participants recruited from the community without renal impairment to undergo contrast-enhanced dual-energy CT. Pulmonary blood volume was calculated by material decomposition; PMBV was defined as blood volume in the peripheral 2 cm of the lung. Non-contrast CT was acquired to assess per cent emphysema and novel CT emphysema subtypes, which include the diffuse emphysema subtype and small-airways-related combined bronchitic-apical emphysema subtype. Generalised linear regression models included age, sex, race/ethnicity, body size, smoking, total lung volume and small airway count. Results Among 495 participants, 53% were never-smokers and the race/ethnic distribution was 35% white, 31% black, 15% Hispanic and 18% Asian. Mean PMBV was 352±120 mL; mean per cent emphysema was 4.95±4.75%. Lower PMBV was associated with greater per cent emphysema (−0.90% per 100 mL PMBV, 95% CI: −1.29 to –0.51). The association was of larger magnitude in participants with 10 or more pack-years smoking and airflow obstruction, but present among participants with no smoking history or airflow limitation, and was specific to the diffuse CT emphysema subtype (−1.48% per 100 mL PMBV, 95% CI: −2.31 to –0.55). Conclusion In this community-based study, lower PMBV was associated with greater per cent emphysema, including in participants without a smoking history or airflow limitation, and was specific to the diffuse CT emphysema subtype. (10.1136/thorax-2024-222002)
  DOI : 10.1136/thorax-2024-222002
• The syzygy distinguisher
  
  • Randriambololona Hugues
  , 2025, 15606, pp.324-354. We present a new distinguisher for alternant and Goppa codes, whose complexity is subexponential in the error-correcting capability, hence better than that of generic decoding algorithms. Moreover it does not suffer from the strong regime limitations of the previous distinguishers or structure recovery algorithms: in particular, it applies to the codes used in the Classic McEliece candidate for postquantum cryptography standardization. The invariants that allow us to distinguish are graded Betti numbers of the homogeneous coordinate ring of a shortening of the dual code. Since its introduction in 1978, this is the first time an analysis of the McEliece cryptosystem breaks the exponential barrier. (10.1007/978-3-031-91095-1_12)
  DOI : 10.1007/978-3-031-91095-1_12
• Active Bipartite Ranking with Smooth Posterior Distributions
  
  • Cheshire James
  • Clémençon Stephan
  , 2025, Volume 258: International Conference on Artificial Intelligence and Statistics. <div><p>In this article, bipartite ranking, a statistical learning problem involved in many applications and widely studied in the passive context, is approached in a much more general active setting than the discrete one previously considered in the literature. While the latter assumes that the conditional distribution is piece wise constant, the framework we develop permits in contrast to deal with continuous conditional distributions, provided that they fulfill a Hölder smoothness constraint. We first show that a naive approach based on discretisation at a uniform level, fixed a priori and consisting in applying next the active strategy designed for the discrete setting generally fails. Instead, we propose a novel algorithm, referred to as smooth-rank and designed for the continuous setting, which aims to minimise the distance between the ROC curve of the estimated ranking rule and the optimal one w.r.t. the sup norm. We show that, for a fixed confidence level ε &gt; 0 and probability δ ∈ (0, 1), smooth-rank is PAC(ε, δ). In addition, we provide a problem dependent upper bound on the expected sampling time of smooth-rank and establish a problem dependent lower bound on the expected sampling time of any PAC(ε, δ) algorithm. Beyond the theoretical analysis carried out, numerical results are presented, providing solid empirical evidence of the performance of the algorithm proposed, which compares favorably with alternative approaches.</p></div>
• Group-robust Machine Unlearning
  
  • de Min Thomas
  • Roy Subhankar
  • Lathuilière Stéphane
  • Ricci Elisa
  • Mancini Massimiliano
  Transactions on Machine Learning Research, [Amherst Massachusetts]: OpenReview.net, 2022, 2025. Machine unlearning is an emerging paradigm to remove the influence of specific training data (i.e., the forget set) from a model while preserving its knowledge of the rest of the data (i.e., the retain set). Previous approaches assume the forget data to be uniformly distributed from all training datapoints. However, if the data to unlearn is dominant in one group, we empirically show that performance for this group degrades, leading to fairness issues. This work tackles the overlooked problem of non-uniformly distributed forget sets, which we call group-robust machine unlearning, by presenting a simple, effective strategy that mitigates the performance loss in dominant groups via sample distribution reweighting. Moreover, we present MIU (Mutual Information-aware Machine Unlearning), the first approach for group robustness in approximate machine unlearning. MIU minimizes the mutual information between model features and group information, achieving unlearning while reducing performance degradation in the dominant group of the forget set. Additionally, MIU exploits sample distribution reweighting and mutual information calibration with the original model to preserve group robustness. We conduct experiments on three datasets and show that MIU outperforms standard methods, achieving unlearning without compromising model robustness. Source code available at https://github.com/tdemin16/group-robust_machine_unlearning. (10.48550/arXiv.2503.09330)
  DOI : 10.48550/arXiv.2503.09330
• Distributed Energy Bank Optimization Towards Outage Aware Sustainable Cellular Networks
  
  • Balakrishnan Ashutosh
  • De Swades
  • Wang Li-Chun
  IEEE Transactions on Sustainable Computing, IEEE, 2025, 10 (3), pp.503-514. Grid connected and solar powered base stations (BSs) acting as distributed energy sources are increasingly becoming a popular solution to mobile operators. These networks experience double stochasticity due to the space-time variations in energy harvest and BS traffic. Hence, accurate and efficient green energy outage estimation in such networks is a challenging task. In this work, we propose a complutationally efficient cooperative energy transfer based distributed energy bank strategy to alleviate green energy outage and design energy sustainable networks. We first develop low-complexity Markovian frameworks to estimate green energy outage in a standalone BS without energy cooperation (WEC) and a multi-BS energy-cooperative (EC) setting, respectively. For the WEC system, we present a computationally efficient three-state discrete time Markovian statistical model, while the multi-BS EC framework is characterized by a two-state Markov model. The energy outage is studied as a function of capital expenditure (CAPEX), manifesting engineering insights from a service provider's perspective. Subsequently for the EC framework, we formulate a CAPEX optimization problem by jointly optimizing the BS cluster size and solar provisioning on individual BSs. Our results demonstrate that the proposed EC framework alleviates the green energy outage significantly, providing computational efficiency gains and CAPEX savings over the state-of-art approaches. (10.1109/TSUSC.2024.3486976)
  DOI : 10.1109/TSUSC.2024.3486976
• VRSurf: Surface Creation from Sparse, Unoriented 3D Strokes
  
  • Sureshkumar Anandhu
  • Parakkat Amal Dev
  • Bonneau Georges-Pierre
  • Hahmann Stefanie
  • Cani Marie-Paule
  Computer Graphics Forum, Wiley, 2025, 44 (2). Although intuitive, sketching a closed 3D shape directly in an immersive environment results in an unordered set of arbitrary strokes, which can be difficult to assemble into a closed surface. We tackle this challenge by introducing VRSurf, a surfacing method inspired by a balloon inflation metaphor: Seeded in the sparse scaffold formed by the strokes, a smooth, closed surface is inflated to progressively interpolate the input strokes, sampled into lists of points. These are treated in a divide-and-conquer manner, which allows for automatically triggering some additional balloon inflation followed by fusion if the current inflation stops due to a detected concavity. While the input strokes are intended to belong to the same smooth 3D shape, our method is robust to coarse VR input and does not require strokes to be aligned. We simply avoid intersecting strokes that might give an inconsistent surface position due to the roughness of the VR drawing. Moreover, no additional topological information is required, and all the user needs to do is specify the initial seeding location for the first balloon. The results show that VRsurf can efficiently generate smooth surfaces that interpolate sparse sets of unoriented strokes. Validation includes a side-by-side comparison with other reconstruction methods on the same input VR sketch. We also check that our solution matches the user's intent by applying it to strokes that were sketched on an existing 3D shape and comparing what we get to the original one. (10.1111/cgf.70071)
  DOI : 10.1111/cgf.70071
• Differentially Private Policy Gradient
  
  • Rio Alexandre
  • Barlier Merwan
  • Colin Igor
  , 2025. Motivated by the increasing deployment of reinforcement learning in the real world, involving a large consumption of personal data, we introduce a differentially private (DP) policy gradient algorithm. We show that, in this setting, the introduction of Differential Privacy can be reduced to the computation of appropriate trust regions, thus avoiding the sacrifice of theoretical properties of the DP-less methods. Therefore, we show that it is possible to find the right trade-off between privacy noise and trust-region size to obtain a performant differentially private policy gradient algorithm. We then outline its performance empirically on various benchmarks. Our results and the complexity of the tasks addressed represent a significant improvement over existing DP algorithms in online RL.
• EmoDynamiX: Emotional Support Dialogue Strategy Prediction by Modelling MiXed Emotions and Discourse Dynamics
  
  • Wan Sky Chenwei
  • Labeau Matthieu
  • Clavel Chloé
  , 2025, pp.1678-1695. Designing emotionally intelligent conversational systems to provide comfort and advice to people experiencing distress is a compelling area of research. Recently, with advancements in large language models (LLMs), end-to-end dialogue agents without explicit strategy prediction steps have become prevalent. However, implicit strategy planning lacks transparency, and recent studies show that LLMs’ inherent preference bias towards certain socio-emotional strategies hinders the delivery of high-quality emotional support. To address this challenge, we propose decoupling strategy prediction from language generation, and introduce a novel dialogue strategy prediction framework, EmoDynamiX, which models the discourse dynamics between user fine-grained emotions and system strategies using a heterogeneous graph for better performance and transparency. Experimental results on two ESC datasets show EmoDynamiX outperforms previous state-of-the-art methods with a significant margin (better proficiency and lower preference bias). Our approach also exhibits better transparency by allowing backtracing of decision making. (10.18653/v1/2025.naacl-long.81)
  DOI : 10.18653/v1/2025.naacl-long.81
• HISTOIRESMORALES: A French Dataset for Assessing Moral Alignment
  
  • Leteno Thibaud
  • Proskurina Irina
  • Gourru Antoine
  • Velcin Julien
  • Laclau Charlotte
  • Metzler Guillaume
  • Gravier Christophe
  , 2025, pp.2590–2612. Aligning language models with human values is crucial, especially as they become more integrated into everyday life. While models are often adapted to user preferences, it is equally important to ensure they align with moral norms and behaviours in real-world social situations. Despite significant progress in languages like English and Chinese, French has seen little attention in this area, leaving a gap in understanding how LLMs handle moral reasoning in this language. To address this gap, we introduce HISTOIRESMORALES, a French dataset derived from MORALSTORIES, created through translation and subsequently refined with the assistance of native speakers to guarantee grammatical accuracy and adaptation to the French cultural context. We also rely on annotations of the moral values within the dataset to ensure their alignment with French norms. HISTOIRESMORALES covers a wide range of social situations, including differences in tipping practices, expressions of honesty in relationships, and responsibilities toward animals. To foster future research, we also conduct preliminary experiments on the alignment of multilingual models on French and English data and the robustness of the alignment. We find that while LLMs are generally aligned with human moral norms by default, they can be easily influenced with user-preference optimization for both moral and immoral data. (10.48550/arXiv.2501.17117)
  DOI : 10.48550/arXiv.2501.17117