INRIA a CCSD electronic archive server
Not a member yet
122212 research outputs found
Sort by
Transmission Operators Optimisation for the Iterative Solution of a Trefftz Method for Time-Harmonic Wave Propagation Problems
No full textWhen solving time-harmonic wave propagation problems in wide domains, memory considerations impose to resort to iterative solvers. Unfortunately, classic numerical methods (as Finite Element or Discontinuous Galerkin) are usually poorly adapted to these algorithms, in opposition to Trefftz methods: a natural preconditioner leads to a contracting system. Nevertheless, such an iterative convergence can be quite slow: inspired by the Domain Decomposition community, we propose to define generalised outgoing/incoming traces (on which relies an Ultra Weak Variational Formulation) by introducing transmission operators on each mesh face, intending to improve the iterative convergence. A generalised Trefftz formulation can then be obtained, and is described in a generic formalism issued from two-fields Friedrichs systems. In particular, these operators (and their matrix representation) are characterised so as to ensure the associated formulation satisfies the original weak-coercivity and contraction properties. Following DDM literature, the transmission operator is searched as an approximation of the Dirichlet-to-Neumann operator, through analytic forms and a data-driven approach, where the function linking local parameters to the operator is searched as a neural network. The iterative convergence of the associated Trefftz methods is then discussed for these choices, allowing to highlight robust gains in terms of iterations to convergence
An Enterprise Marketplace for Unified Access to Multi-Cloud and Enterprise Products in a Large Banking Infrastructure
No full textInternational audienceWe present the design and evaluation of an Enterprise Marketplace that unifies web-based access to multi-cloud and enterprise products within a large banking infrastructure. Building such a platform poses significant technical and organizational challenges, including the generalization of diverse APIs and the accommodation of heterogeneous user profiles. Our solution enables autonomous product publishing via a no-code interface, enforces multi-layered governance to ensure security and compliance, and integrates disparate providers through a standardized API. We demonstrate how this approach enhances product quality, producer autonomy, and user experience, supported by adoption metrics and operational data from a real-world deployment. Finally, we reflect on key lessons learned and persistent challenges after a decade of production use, serving over 200,000 users
Unlearning Personal Data from a Single Image
No full textInternational audiencePublished in Transactions on Machine Learning Research (TMLR
Morphology-enhanced CAM-guided SAM for weakly supervised breast lesion segmentation
No full textInternational audienceUltrasound imaging is vital for the early detection of breast cancer, where accurate lesion segmentation supports clinical diagnosis and treatment planning. However, existing deep learning-based methods rely on pixel-level annotations, which are costly and laborintensive to obtain. This study presents a weakly supervised framework for breast lesion segmentation in ultrasound images. The framework combines morphological enhancement with Class Activation Map (CAM)-guided lesion localization and utilizes the Segment Anything Model (SAM) for refined segmentation without pixel-level labels. By adopting a lightweight region synthesis strategy and relying solely on SAM inference, the proposed approach substantially reduces model complexity and computational cost while maintaining high segmentation accuracy. Experimental results on the BUSI dataset show that our method achieves a Dice coefficient of 0.7063 under five-fold cross-validation and outperforms several fully supervised models in Hausdorff distance metrics. These results demonstrate that the proposed framework effectively balances segmentation accuracy, computational efficiency, and annotation cost, offering a practical and low-complexity solution for breast ultrasound analysis. The code for this study is available at: https://github.com/YueXin18/MorSeg-CAM-SAM-Segmentation
Complete Abstractions for Verification of Polymorphic Functions with Equality -- extended version
No full textThis paper is concerned with automatically proving properties on polymorphic programs over algebraic data types by reducing the verification of such properties to the verification of properties on monomorphised, abstract programs. For programs without polymorphic equality, the reduction exploits Wadler's "Theorem for Free". For programs using polymorphic equality, we provide a sufficient condition for the reduction to hold. The condition relies on the existence of a locally complete abstraction function whose image is a finite set of arbitrary constants chosen for abstracting primitive values. When such a condition exists, the number of arbitrary constants depends on the functions under concern and the properties to prove. We present an implementation that automatically computes the number of constants and, thus, ensures that proving the polymorphic case with equality can be reduced to the proof carried out on a monomorphic instance of the program. Experimental results show that this reduction is indeed possible with small abstract domains. Target programs support user-defined recursive ADTs and recursive first-order functions.</div
S4 modal sequent calculus as intermediate logic and intermediate language
No full textInternational audienceIn this short paper, we advocate for the idea that continuation-based intermediate languages correspondto intermediate logics. The goal of intermediate languages is to serve as a basis for compiler intermediaterepresentations, allowing to represent expressive program transformations for optimisation and compilation,while preserving the properties that make programs compilable efficiently in the first place, such as the“stackability” of continuations. Intermediate logics are logics between intuitionistic and classical logic in termsof provability.Second-class continuations used in CPS-based intermediate languages correspond to a classical modal logicS4 with the added restriction that implications may only return modal types. This indeed corresponds to anintermediate logic, owing to the Gödel-McKinsey-Tarski theorem which states the intuitionistic nature of themodal fragment of S4.We introduce a three-kinded polarised sequent calculus for S4, together with an operational machinemodel that separates a heap from a stack. With this model we study a stackability property for the modalfragment of S4
Evolution of multicellular reproduction through co-option of ecological interactions
No full textDuring the transition to multicellularity, cells evolved novel regulatory mechanisms to coordinate cell division and differentiation, which enabled the emergence of group reproduction. These mechanisms were repurposed from molecular and cellular traits that once mediated interactions between single-celled organisms. However, it remains unclear how these traits were evolutionarily integrated to form the first developmental programs in multicellular life. To address this issue, we developed a spatially structured evolutionary model in which cells can migrate, divide, and adhere to their neighbors — behaviors common to most unicellular eukaryotes. When coupled to a selective pressure driven by food scarcity, the model reveals that the ecological context plays a central role in the evolution of multi-cellular reproduction. Depending on the spatial distribution of food in the environment, both unicellular and multicellular life cycles with diverse modes of reproduction could evolve. Among these were multicellular life cycles that reproduce through unicellular propagules, the most prevalent reproductive strategy in multicellular life, which emerged spontaneously as a dispersal strategy in some environments. We show that these propagules are genetically homologous to the lineage’s unicellular ancestors, which were co-opted and repurposed as reproductive structures during evolution. Furthermore, after multicellular lineages with propagules evolved, they could colonize environments that were previously dominated by unicellular life. Altogether, our results show how ecological interactions between single cells can transform into developmental processes during the evolutionary transition towards multicellularity. Significance statement Reproduction is a universal feature of life. Yet, the evolution of multicellularity transformed it fundamentally: while single-celled organisms reproduce via cell division, reproduction in multicellular organisms is a complex process involving the coordination of many cells. How these new forms of multicellular reproduction first evolved is currently unknown. Using a computational model, we study how group reproduction emerges from the collective dynamics of individual cells. The model shows that unicellular ancestral life cycles can be repurposed as propagules used for reproduction in multicellular species, suggesting that genetic co-option is a key mechanism through which early development evolves
Hétérodimérisation du récepteur aux oestrogènes couplés aux protéines G avec le récepteur de la LH biaise la signalisation dépendante des proteines G
No full textInternational audienceLH/choriogonadotropin (hCG) receptor (LHCGR) and the G protein-coupled estrogen receptor (GPER) are coexpressed in the ovary and support reproduction. The latter is involved in pathophysiological conditions and has been suggested as a potential therapeutic target. However, its role is still controversial, and several studies reported GPER to form heterocomplexes with other class A G protein-coupled receptors, modulating their signaling cascades. We evaluated if GPER interacts with LHCGR and impacts ligand-mediated pathways. In HEK293, LHCGR-GPER heteromers allosterically modulate LH/hCG-mediated signaling by preventing receptor coupling with Gq protein, leading to inhibition of phospholipase C pathway, and related transcriptional and mitogenic functions. This effect is prevented by mutant GPER unable to form heteromers with LHCGR. Interestingly, GPER expression has no effect on LH/hCG-induced Gs/cAMP/protein kinase A pathway activation, demonstrating selective inhibition of Gq pathway. These results were not recapitulated in cells displaying insufficient endogenous Gq protein expression levels, whereas they are recovered under exogenous Gq overexpression. Our data strengthen the concept that GPER may act as a modulator of other membrane G protein-coupled receptors, and a potential new target for treatment of tumors displaying Gq signalling
Evaluating portable EEG: a comparison between two wireless systems (EPOC Flex and LiveAmp) and the wired BrainAmp system
No full textInternational audienceBackground Recent advances in equipment miniaturization have led to low-cost, portable electroencephalography (EEG) systems that facilitate data collection in real-world settings and with larger samples. Although wireless EEG systems were originally developed for non-research applications, recent studies have provided valuable information to help researchers make informed choices, particularly about participant comfort, mobility during recordings, and data validity. This study aimed to assess the impact of portability by comparing the performance of portable consumer- and research-grade systems (EPOC Saline Flex, EM; LiveAmp, LA) with fixed research-grade systems (BrainAmp, BA). Method Continuous EEG was recorded with each system in healthy adults performing five benchmark tasks in fundamental and clinical cognitive neuroscience. Mental states (alpha power variations in open/closed eyes) and unconscious perception (steady-state visual evoked potential, SSVEP) were analyzed through time/frequency methods, while active (N200 and P300 components during active listening and N170 component during face recognition) and passive cognitive processes (Mismatch negativity, MMN component during passive listening) were examined using time/amplitude analyses (event-related potential, ERPs). Our analyses compared system efficiency at native and equalized sampling rates and examined 100%, 75%, and 50% of the datasets to determine the required trial number for satisfactory signal quality. Results Despite the smaller amount of signal retained for EM, all systems recorded the expected resting state alpha power decrease and SSVEP responses, with EM showing lower spectral effects ([EM < (LA≈BA)]). ERPs for active (N170, N200, P300) and passive (MMN) processes emerged across all systems, with EM and LA showing lower amplitudes only for N170 compared to BA. Furthermore, the dataset reduction resulted in a decreased N170 at P7 only for EM ([EM < LA < BA]). EM also exhibited shorter latencies for all ERPs except for MMN. Conclusion This study provides concrete guidance for designing EEG experiments in real-world settings, with significant potential for investigating children and vulnerable populations. The efficiency of the three EEG systems is more influenced by task duration than sampling rates. A wireless EEG device, such as the EM, can effectively support both time/frequency and time/amplitude analyses in cognitive science, provided that the number of trials is sufficient and latencies are controlled
Attacking the First-Principle: A Black-Box, Query-Free Targeted Mimicry Attack on Binary Function Classifiers
No full textBinary function classifiers play a crucial role in maintaining the security and integrity of software systems by detecting malicious code and unauthorized modifications. However, machine learning-based classifiers are vulnerable to adversarial attacks that can evade detection. In this study, we present Kelpie, a novel framework for executing mimicry attacks, a stronger type of targeted evasion attacks, on binary function classifiers in a black-box, zero-query setting. Unlike previous approaches that rely on querying the target classifier to refine untargeted evasion attacks, Kelpie leverages code transformations that preserve the functionality of malicious payloads while causing them to be misclassified as we want. Through extensive experimentation, we demonstrate that Kelpie can successfully execute mimicry attacks against six state-of-the-art binary function classifiers representing different model architectures without requiring direct interaction with them. We further validate our approach with a practical demonstration, involving a keylogger and a wiper concealed within benign-looking functions embedded in an application. This work, to our best knowledge, is the first to demonstrate such a mimicry attack in a black-box, zero-query context, raising important questions about the reliability and security of existing machine learning-based binary function classifiers