INRIA a CCSD electronic archive server
Not a member yet
122212 research outputs found
Sort by
Le plan corporel des animaux revisité à la lumière de l'échec de la cohésion tissulaire locale dans le cancer
No full textArticle soumis à un journalThe body plan of multicellular animals is here revisited as the program of construction, contained in the zygote as a self-extracting archive -and later borne in all nucleated cells of any given multicellular animal -which is launched at fecundation. It unfolds in embryogenesis by a succession of differentiation branchings in the cellular matter produced by successive cell proliferations, leading to the different cell types a multicellular organism -in its achieved form -is made of, 20 for sponges (Porifera), 200 to 400 for Humans. The body plan, once has been physically achieved as coherent and viable the animal multicellular organism for which it is designed, must be maintained in the different constituting tissues and organs of the organism by local cohesion control mechanisms, some of which might be linked to the permanent action of tissue resident macrophages, already present from early embryogenesis. Cancer is a disease of animal multicellular organisms only, and its beginning always occurs in a given tissue, in which some such cohesion controls fail, firstly on differentiation, secondly on proliferation. This essay theoretically investigates possible control mechanisms of tissue cohesion maintenance failed in cancer, attempting in particular at eliciting a role for resident macrophages in such local tissue maintenance
Toward Valid Generative Clinical Trial Data with Survival Endpoints
No full textInternational audienceClinical trials face mounting challenges: fragmented patient populations, slow enrollment, and unsustainable costs, particularly for late-phase trials in oncology and rare diseases. While external control arms built from real-world data have been explored, a promising alternative is the generation of synthetic control arms using generative AI. A central challenge is the generation of time-to-event outcomes, which constitute primary endpoints in oncology and rare disease trials, but are difficult to model under censoring and small sample sizes. Existing generative approaches, largely GAN-based, are data-hungry, unstable, and rely on strong assumptions such as independent censoring. We introduce a variational autoencoder (VAE) that jointly generates mixed-type covariates and survival outcomes within a unified latent variable framework, without assuming independent censoring. Across synthetic and real trial datasets, we evaluate our model in two realistic scenarios: (i) data sharing under privacy constraints, where synthetic controls substitute for original data, and (ii) control-arm augmentation, where synthetic patients mitigate imbalances between treated and control groups. Our method outperforms GAN baselines on fidelity, utility, and privacy metrics, while revealing systematic miscalibration of type I error and power. We propose a post-generation selection procedure that improves calibration, highlighting both progress and open challenges for generative survival modeling
Methods for a species-specific genome-scale metabolic model designed for eukaryotes and applied to the Ascophyllum nodosum macroalga
No full textNational audienceGenome-scale metabolic models (GEMs) are essential tools for studying metabolism, either for comparative analyses or to investigate interactions between organisms. However, genome annotation, biomass formulation, and network gap-filling are key steps in constructing a relevant GEM and ensuring the biosynthesis of specialized metabolites. We present a pipeline to integrate extensive biological knowledge (genomes of closely related species, metabolic profiling studies, potential interactions with microbiota) about an eukaryotic organism in order to generate high quality GEMs. To manage genome annotation limitations, the pipeline relies on a GEM reconstruction tool that propagates annotations across closely related species through the identification of orthologous genes. It also pays particular attention to biomass formulation, using a set of metabolomic studies to create a consensus biomass composition that seeks to closely reflect biological reality, such as incorporating specialized metabolites and their precursors. The gap-filling stage of the pipeline uses a semi-automated curation process for added reactions, taking into account the presence of orthologous genes, occurrence in phylogenetically related species and potential interactions with the organism's microbiota. The final GEM applied to the brown alga Ascophyllum nodosum comprises 3,536 metabolites and 3,072 biochemical reactions, predicting the synthesis of 1,023 compounds from 38 seawater-derived metabolites. Almost all reactions (99.98%) are linked to an enzyme supported in the algal genome. This refined model provides a framework for studying host-microbiota metabolic complementarity. This pipeline offers a scalable and robust method for reconstructing high-quality GEMs in emerging eukaryotic model organisms, improving metabolic network accuracy and expanding our understanding of species-specific metabolism. It also sheds lights on the various level of knowledge related to the synthesis pathways of the biomass, paving the way to future studies to be undergone
MAP Estimation with Denoisers: Convergence Rates and Guarantees
No full textInternational audienceDenoiser models have become powerful tools for inverse problems, enabling the use of pretrained networks to approximate the score of a smoothed prior distribution. These models are often used in heuristic iterative schemes aimed at solving Maximum a Posteriori (MAP) optimisation problems, where the proximal operator of the negative log-prior plays a central role. In practice, this operator is intractable, and practitioners plug in a pretrained denoiser as a surrogate-despite the lack of general theoretical justification for this substitution. In this work, we show that a simple algorithm, closely related to several used in practice, provably converges to the proximal operator under a log-concavity assumption on the prior p. We show that this algorithm can be interpreted as a gradient descent on smoothed proximal objectives. Our analysis thus provides a theoretical foundation for a class of empirically successful but previously heuristic methods
Stylized Meta-Album: Group-bias injection with style transfer to study robustness against distribution shifts
No full textWe introduce Stylized Meta-Album (SMA), a new image classification meta-dataset comprising 24 datasets (12 content datasets, and 12 stylized datasets), designed to advance studies on out-of-distribution (OOD) generalization and related topics. Created using style transfer techniques from 12 subject classification datasets, SMA provides a diverse and extensive set of 4800 groups, combining various subjects (objects, plants, animals, human actions, textures) with multiple styles. SMA enables flexible control over groups and classes, allowing us to configure datasets to reflect diverse benchmark scenarios. While ideally, data collection would capture extensive group diversity, practical constraints often make this infeasible. SMA addresses this by enabling large and configurable group structures through flexible control over styles, subject classes, and domains—allowing datasets to reflect a wide range of real-world benchmark scenarios. This design not only expands group and class diversity, but also opens new methodological directions for evaluating model performance across diverse group and domain configurations—including scenarios with many minority groups, varying group imbalance, and complex domain shifts—and for studying fairness, robustness, and adaptation under a broader range of realistic conditions. To demonstrate SMA's effectiveness, we implemented two benchmarks: (1) a novel OOD generalization and group fairness benchmark leveraging SMA's domain, class, and group diversity to evaluate existing benchmarks. Our findings reveal that while simple balancing and algorithms utilizing group information remain competitive as claimed in previous benchmarks, increasing group diversity significantly impacts fairness, altering the superiority and relative rankings of algorithms. We also propose to use \textit{Top-M worst group accuracy} as a new hyperparameter tuning metric, demonstrating broader fairness during optimization and delivering better final worst-group accuracy for larger group diversity. (2) An unsupervised domain adaptation (UDA) benchmark utilizing SMA's group diversity to evaluate UDA algorithms across more scenarios, offering a more comprehensive benchmark with lower error bars (reduced by 73\% and 28\% in closed-set setting and UniDA setting, respectively) compared to existing efforts. These use cases highlight SMA's potential to significantly impact the outcomes of conventional benchmarks
Reliable and Robust Watermarking for Data Flooding against Ransomware Random Techniques
No full textInternational audienceData Flooding Against Ransomware (DFaR) techniques combat ransomware through decoy files that can reveal a ransomware's activity and reduce the effectiveness and efficiency of attacks by confounding legitimate user files and competing for IO resource access of the attacked host. While effective, existing DFaR random strategies (which flood a user system with realistic yet random-content decoy files) face challenges during restoration, due to the necessity of pre-attack file lists to discriminate between proper and decoy files (the latter should be removed to restore the system to its pre-attack state). To tackle this issue, we present a watermarking-based approach that embeds imperceptible watermarks in random-content decoy files. Our technique preserves the indistinguishability of decoys from user files to attackers, while providing users with a reliable mechanism to differentiate between authentic and decoy content, obviating the need for pre-attack file lists. We present experimental evaluations that demonstrate that our watermarking technique a) imposes minimal-to-medium computational overhead (depending on user-configurable parameters) compared to existing random-content flooding methods (i.e., it is efficient when contrasting ransomware and restoring a user's system) and b) it provides strong resistance against adversarial inference attacks
Un code non monotone de la probabilité d'un événement dans le cerveau humain
No full textInternational audienceAssessing probabilities and predicting future events are fundamental for perception and adaptive behavior, yet the neural representations of probability remain elusive. While previous studies have shown that neural activity in several brain regions correlates with probability-related factors such as surprise and uncertainty, similar correlations have not been found for probability. Here, using 7 Tesla functional magnetic resonance imaging, we uncover a representation of the probability of the next event in a sequence within the human dorsolateral prefrontal and intraparietal cortices. Crucially, univariate and multivariate analyses revealed that this representation employs a highly non-monotonic code. Tuning curves for probability exhibit selectivity to various probability ranges, while the code for confidence accompanying these estimates is predominantly monotonic. Given such diversity in tuning curves, future studies should move from assuming monotonic or simple canonical forms of tuning curves to considering richer representations, and clarify why different types of code exist.L'évaluation des probabilités et la prédiction des événements futurs sont fondamentales pour la perception et le comportement adaptatif, mais les représentations neuronales de la probabilité restent méconnues. Si des études antérieures ont montré que l'activité neuronale dans plusieurs régions du cerveau est corrélée à des facteurs liés à la probabilité, tels que la surprise et l'incertitude, aucune corrélation similaire n'a été trouvée pour la probabilité. Dans cette étude, à l'aide d'une imagerie par résonance magnétique fonctionnelle 7 Tesla, nous décrivons une représentation de la probabilité du prochain événement dans une séquence au sein des cortex préfrontal dorsolatéral et intrapariétal humains. Des analyses univariées et multivariées ont révélé que cette représentation utilise un code hautement non monotone. Les courbes de réponse à la probabilité présentent une sélectivité pour différentes plages de probabilité, tandis que le code de confiance accompagnant ces estimations est principalement monotone. Compte tenu de cette diversité dans les courbes de réponse, les études futures devraient passer de l'hypothèse de courbes de réponse monotones ou de formes canoniques simples à la prise en compte de représentations plus riches, et clarifier pourquoi différents types de codes existent
Decoupling actions of finite-dimensional Lie groups and of groups of diffeomorphisms in the large deformation framework
No full textIn computational anatomy, the Large Deformation Diffeomorphic Metric Mapping (LDDMM) framework has become a central tool for modeling smooth, invertible transformations between shapes such as curves or landmarks. In this paper, we extend this framework by enriching diffeomorphic deformations with transformations induced by finite-dimensional Lie groups (e.g. isometries, scalings), and we develop a registration model that decouples the actions of these two types of deformation on the shape during the matching process. To achieve this, we consider semidirect products between finite-dimensional groups and groups of diffeomorphisms, endowed with a right-invariant sub-Riemannian structure that give rise to new variational problems for shape registration. By exploiting symmetries and reduction theory, we decouple the contributions of each group throughout the matching process. We further extend the framework to incoroporate anisotropic deformations that preferentially favor certain directions during registration. On the numerical side, we propose an algorithm based on a joint optimization over both deformation groups, in contrast to the standard twostage approach that optimizes first over the finite-dimensional component and then over the diffeomorphic one. Experiments on curves and landmarks demonstrate that the proposed joint optimization improves registration accuracy and more effectively disentangles the contributions of the two deformation groups
Optimal allocation control in microbial growth under a heat-shock
No full textsubmitted to ECC 2026Resource allocation models are a highly useful tool to study microbial growth in a variety of contexts. We use here a model of this kind describing the metabolism of a micro-organism subjected to a heat-shock. Under extreme temperatures, proteins have a higher tendency to unfold and lose their ability to perform their function, which is counteracted by an increase in the production of folding-assisting chaperones. The model considers a two-dimensional control, representing precursors allocation to the production of chaperones and ribosomes, with the remainder going to the metabolic machinery. We are interested in determining the control that maximizes growth rate following a temperature change. We begin with a theoretical analysis of the resulting optimal control problem (OCP) by use of Pontryagin's Maximum Principle (PMP). We show that this control follows a bang-bang structure, and characterize and give analytical expressions for some of the singular arcs that can emerge. We use the numerical resolution package OptimalControl.jl to solve the OCP in fixed final time in the case of a heat-shock from 37°C to 42 °C. The resulting control presents a simple structure consisting of an arc maximizing chaperone production followed by a constant arc corresponding to the optimal equilibrium in the different internal concentrations at the new temperature, in line with biological observations. Finally, we show how the optimal response to a heat-shock follows this structure independently of the temperature prior to the shock