INRIA a CCSD electronic archive server
Not a member yet
122212 research outputs found
Sort by
MetaNetMap : cartographie automatique des données métabolomiques sur les réseaux métaboliques
No full textInternational audienceMetabolic networks represent genome-derived information about the biochemical reactions that cells are capable of performing. Mapping omic data onto these networks is important to refine model simulations. However, metabolomic data mapping remains very challenging due to difficulties in identifier reconciliation between annotation profiles and metabolic networks. MetaNetMap is a Python package designed to automatise the process of mapping metabolomic data onto metabolic networks. It includes several layers of identifier matching, the use of customisable databases, and molecular ontology integration to suggest the most matches between experimentally-identified metabolites and molecules defined in the network.Les réseaux métaboliques représentent les informations issues du génome concernant les réactions biochimiques que les cellules sont capables d'effectuer. La cartographie des données omiques sur ces réseaux est importante pour affiner les simulations de modèles. Cependant, la cartographie des données métabolomiques reste très difficile en raison des difficultés de rapprochement des identifiants entre les profils d'annotation et les réseaux métaboliques. MetaNetMap est un package Python conçu pour automatiser le processus de cartographie des données métabolomiques sur les réseaux métaboliques. Il comprend plusieurs niveaux de correspondance des identifiants, l'utilisation de bases de données personnalisables et l'intégration de l'ontologie moléculaire afin de suggérer les correspondances les plus pertinentes entre les métabolites identifiés expérimentalement et les molécules définies dans le réseau
Applying the maximum entropy principle to neural networks enhances multi-species distribution models
No full textSubmitted to Methods in Ecology and EvolutionThe rapid expansion of citizen science initiatives has led to a significant growth of biodiversitydatabases, and particularly presence-only (PO) observations. PO data are invaluable for understanding species distributions and their dynamics, but their use in a Species Distribution Model (SDM) is curtailed by sampling biases and the lack of information on absences. Poisson point processes are widely used for SDMs, with Maxent being one of the most popular methods. Maxent maximises the entropy of a probability distribution across sites as a function of predefined transformations of variables, called features. In contrast, neural networks and deep learning have emerged as a promising technique for automatic feature extraction from complex input variables. Arbitrarily complex transformations of input variables can be learned from the data efficiently through backpropagation and stochastic gradient descent (SGD). Yet, deep learning was mainly developed for classification problems, and learning robust features and species abundances across space while properly correcting for sampling biases has remained a challenge so far. In this paper, we propose DeepMaxent, which harnesses neural networks to automatically learn shared features among species, using the maximum entropy principle. To do so, it employs a normalised Poisson loss where for each species, presence probabilities across sites are modelled by a neural network. We evaluate DeepMaxent on a benchmark dataset known for its spatial sampling biases, using PO data for calibration and presence-absence (PA) data for validation across six regions with different biological groups and covariates. Our results indicate that DeepMaxent performs better than Maxent and other leading SDMs across all regions and taxonomic groups. The method performs particularly well in regions of uneven sampling, demonstrating substantial potential to increase SDM performances. The method opens the possibility to learn more robust features predicting simultaneously many species to arbitrary large datasets without increased memory requirements. The model likelihood, arising from a Poisson process, makes the method compatible with the integration of more standardised types of data to further increase sampling bias correction. In particular, our approach yields more accurate predictions than traditional single-species models, which opens up new possibilities for methodological enhancement
Revisiting PQ WireGuard: A Comprehensive Security Analysis With a New Design Using Reinforced KEMs
No full textInternational audienceWireGuard is a VPN based on the Noise protocol, known for its high performance, small code base, and unique security features. Recently, Hülsing et al. (IEEE S&P'21) presented post-quantum (PQ) WireGuard, replacing the Diffie-Hellman (DH) key exchange underlying the Noise protocol with key-encapsulation mechanisms (KEMs). Since WireGuard requires the handshake message to fit in one UDP packet of size roughly 1200 B, they combined Classic McEliece and a modified variant of Saber. However, as Classic McEliece public keys are notoriously large, this comes at the cost of severely increasing the server's memory requirement. This hinders deployment, especially in environments with constraints on memory (allocation), such as a kernel-level implementations.In this work, we revisit PQ WireGuard and improve it on three fronts: design, (computational) security, and efficiency. As KEMs are semantically, but not syntactically, the same as DH key exchange, there are many (in hindsight) ad-hoc design choices being made, further amplified by the recent finding on the binding issues with PQ KEMs (Cremers et al., CCS'24). We redesign PQ WireGuard addressing these issues, and prove it secure in a new computational model by fixing and capturing new security features that were not modeled by Hülsing et al. We further propose 'reinforced KEM' (RKEM) as a natural building block for key exchange protocols, enabling a PQ WireGuard construction where the server no longer needs to store Classical McEliece keys, reducing public key memory by 190 to 390×. In essence, we construct a RKEM named 'Rebar' to compress two ML-KEM-like ciphertexts which may be of an independent interest
The Strong (2,2)-Conjecture for more classes of graphs
No full textInternational audienceThe Strong -Conjecture asks whether, for all connected graphs different from and , we can assign to edges red and blue labels with value or so that no two adjacent vertices have the same sum of incident red labels or the same sum of incident blue labels. This conjecture, which can be perceived as a generalisation of the so-called 1-2-3 Conjecture, as, thus far, been proved only for a handful number of graph classes. In this work, we prove the Strong -Conjecture holds for more classes of graphs. In particular, we prove the conjecture for cacti, subcubic outerplanar graphs, graphs with maximum average degree less than , and some Halin graphs, among others
DIVA: An Ontology-based Approach to Model User Activity within Visualization Systems
No full textInternational audienceThe study of user activity supports evaluation of visualization systems, recommendation of suitable views or tasks, guidance of interaction, and validation of analytical results. It enables researchers to understand how these visualization systems are used and to gain insight into users' reasoning processes during data exploration. However, there is a lack of structured frameworks for systematically collecting and reasoning over such data. In this paper, we build upon Semantic Web standards to model and represent user activity as knowledge graphs. We introduce an OWL ontology specifically designed for this purpose and demonstrate its application by transforming system log data-collected during user studies with a multiview visualization tool for urban mobility data exploration-into an RDF knowledge graph. Finally, we illustrate the utility and expressiveness of our model by enabling intuitive exploration and interpretation of user activity through the implementation of competency questions as SPARQL queries and the visualization of these queries' results on the RDF graph representing user activity
Reclaiming Software Engineering as the Enabling Technology for the Digital Age
No full textInternational audienceSoftware engineering is the invisible infrastructure of the digital age. Every breakthrough in artificial intelligence, quantum computing, photonics, and cybersecurity relies on advances in software engineering, yet the field is too often treated as a supportive digital component rather than as a strategic, enabling discipline. In policy frameworks, including major European programmes, software appears primarily as a building block within other technologies, while the scientific discipline of software engineering remains largely absent. This position paper argues that the long-term sustainability, dependability, and sovereignty of digital technologies depend on investment in software engineering research. It is a call to reclaim the identity of software engineering
A Year Under the DSA: Ad Transparency's Uneven Landscape
No full textInternational audienceThe Digital Services Act (DSA) has put platform accountability on center stage, requiring online platforms to provide greater transparency into how advertisements are targeted and delivered to users. Central to these obligations are two mechanisms: user-facing ad explanations, which inform individuals why they were shown a given ad, and public ad repositories, which are intended to enable independent auditing of advertising practices. This study provides the first multi-platform evaluation of these two mechanisms across Facebook, Instagram, YouTube and X. Using 48,511 user-facing "Why am I seeing this ad?" (WAIST) notices, and a systematic analysis of each platform's public ad repository, we assess how well current implementations disclose the parameters and decision processes involved in targeting. To do so, we develop and apply an operational framework based on Articles 26 and 39 of the DSA-capturing the granularity, attribution of targeting and delivery choices, data source disclosures, and accuracy-and apply it across both user-facing notices and public ad repositories. Our findings show that transparency remains fragmented and inconsistent across platforms. User-facing explanations vary widely in precision and often omit key targeting information, while repositories provide incomplete, misattributed, and at times difficult-to-interpret targeting data. Moreover, discrepancies between explanations and repository entries undermine the reliability of both mechanisms. Overall, current transparency infrastructures fall short of the DSA's expectations and highlight the need for clearer and more enforceable standards for advertising transparency moving forward.</div
Centered colorings in minor-closed graph classes
No full textInternational audienceA vertex coloring of a graph is -centered if for every connected subgraph of , either uses more than colors on , or there is a color that appears exactly once on . We prove that for every fixed positive integer , every -minor-free graph admits a -centered coloring using colors
Domain Adaptation with a Single Vision-Language Embedding
No full textUnder reviewInternational audienceDomain adaptation has been extensively investigated in computer vision but still requires access to target data at the training time, which might be difficult to obtain in some uncommon conditions. In this paper, we present a new framework for domain adaptation relying on a single Vision-Language (VL) latent embedding instead of full target data. First, leveraging a contrastive language-image pre-training model (CLIP), we propose prompt/photo-driven instance normalization (PIN). PIN is a feature augmentation method that mines multiple visual styles using a single target VL latent embedding, by optimizing affine transformations of low-level source features. The VL embedding can come from a language prompt describing the target domain, a partially optimized language prompt, or a single unlabeled target image. Second, we show that these mined styles (i.e., augmentations) can be used for zero-shot (i.e., target-free) and one-shot unsupervised domain adaptation. Experiments on semantic segmentation demonstrate the effectiveness of the proposed method, which outperforms relevant baselines in the zero-shot and one-shot settings