HAL-CentraleSupelec
Not a member yet
77624 research outputs found
Sort by
Multiphysics uncertainty decoupling methods for stochastic poroelastic analysis
No full textInternational audienc
Meta-DAN: towards an efficient prediction strategy for page-level handwritten text recognition
No full textInternational audienceRecent advances in text recognition led to a paradigm shift for page-level recognition, from multi-step segmentation-based approaches to end-to-end attention-based ones. However, the naïve character-level autoregressive decoding process results in long prediction times: it requires several seconds to process a single page image on a modern GPU. We propose the Meta Document Attention Network (Meta-DAN) as a novel decoding strategy to reduce the prediction time while enabling a better context modeling. It relies on two main components: windowed queries, to process several transformer queries altogether, enlarging the context modeling with near future; and multi-token predictions, whose goal is to predict several tokens per query instead of only the next one. We evaluate the proposed approach on 10 full-page handwritten datasets and demonstrate state-of-the-art results on average in terms of character error rate. Source code and weights of trained models are available at https://github.com/FactoDeepLearning/meta_dan
Area Efficient Speculative Loop Pipelining for High-Level Synthesis
No full textInternational audienceHigh-Level Synthesis (HLS) allows the automatic generation of efficient circuit designs for computation-intensive kernels, but it lacks flexibility when dealing with irregular control flow. Dynamic and speculative HLS techniques are used to address this issue. These techniques outperform state-of-the-art HLS in kernel execution times but introduce a significant area overhead. In contrast, state-of-the-art HLS easily highlights and exploits resource-sharing opportunities. In this work, we show how to adapt an existing speculative HLS approach to take advantage of well-known static resource sharing mechanisms. Our results show a decrease of the area cost by 34% on average
DUALF-D: Disentangled dual-hyperprior approach for light field image compression
No full textInternational audienceLight field (LF) imaging captures spatial and angular information, offering a 4D scene representation enabling enhanced visual understanding. However, high dimensionality and redundancy across spatial and angular domains present major challenges for compression, particularly where storage, transmission bandwidth, or processing latency are constrained. We present a novel Variational Autoencoder (VAE)-based framework that explicitly disentangles spatial and angular features using two parallel latent branches. Each branch is coupled with an independent hyperprior model, allowing more precise distribution estimation for entropy coding and finer rate-distortion control. This dual-hyperprior structure enables the network to adaptively compress spatial and angular information based on their unique statistical characteristics, improving coding efficiency. To further enhance latent feature specialization and promote disentanglement, we introduce a mutual information-based regularization term that minimizes redundancy between the two branches while preserving feature diversity. Unlike prior methods relying on covariance-based penalties prone to collapse, our information-theoretic regularizer provides more stable and interpretable latent separation. Experimental results on publicly available LF datasets demonstrate our method achieves strong compression performance, yielding an average BD-PSNR gain of 2.91 dB over HEVC and high compression ratios (e.g., 200:1). Additionally, our design enables fast inference, with a total end-to-end time over 19x faster than the JPEG Pleno standard, making it well-suited for real-time and bandwidth-sensitive applications. By jointly leveraging disentangled representation learning, dual-hyperprior modeling, and information-theoretic regularization, our approach offers a scalable, effective solution for practical light field image compression.</div
Automatic Extraction of Timing Models for WCET Estimation From a High-Level Synthesis Flow
No full textInternational audienceReal-time, domain-specific processors require faithful timing models for WCET analysis. However, existing models are typically hand-crafted from sparse documentation, making them error-prone and difficult to maintain. This work aims to automatically extract WCET timing models from single-issue in-order processor pipelines generated by High-Level Synthesis (HLS). By deriving timing models directly from the SpecHLS intermediate representation, the models are faithful by construction. Experimental results show that our timing-model extraction process generalizes across diverse RISC-V core variants and yields WCET estimates within 0.48% on average of those from a handcrafted model, on the Mälardalen WCET benchmarks
Single-cell exploration of gonadal somatic cell lineage specification during human sex determination
No full textInternational audienceGonad development is an exciting model to study cell fate commitment. A better understanding of sex determination requires the identification of all involved cell types and their dynamic expression programs. Here we present an atlas of 128,000 single cells from human gonads between 5 and 12 postconceptional weeks. A focused analysis of somatic cells uncovered a population of bipotential progenitors derived from the coelomic epithelium of both testes and ovaries, which may have the capacity to commit to either a steroidogenic or a supporting fate. Moreover, our analyses suggest that early supporting cells, prior to differentiation into Sertoli or pre-granulosa cells, also give rise to the rete testis/ovarii and that the ovary retains the capacity to feed the supporting cell pool for an extended period of time, directly from the surface epithelium. Finally, the potential involvement of the GnRH signaling pathway in regulating testis differentiation was assessed ex vivo
Backstepping for partial differential equations: A survey
No full textInternational audienceSystems modeled by partial differential equations (PDEs) are at least as ubiquitous as those by nature finite-dimensional and modeled by ordinary differential equations (ODEs). And yet, systematic and readily usable methodologies, for such a significant portion of real systems, have been historically scarce. Around the year 2000, the backstepping approach to PDE control began to offer not only a less abstract alternative to PDE control techniques replicating optimal and spectrum assignment techniques of the 1960s, but also enabled the methodologies of adaptive and nonlinear control, matured in the 1980s and 1990s, to be extended from ODEs to PDEs, allowing feedback synthesis for systems that are uncertain, nonlinear, and infinite-dimensional. The PDE backstepping literature has since grown to hundreds of papers and nearly a dozen books. This survey aims to facilitate the entry into this thriving area of overwhelming size and topical diversity. Designs of controllers and observers, for parabolic, hyperbolic, and other classes of PDEs, in one or more dimensions, with nonlinear, adaptive, sampled-data, and event-triggered extensions, are covered in the survey. The lifeblood of control are technology and physics. The survey places a particular emphasis on applications that have motivated the development of the theory and which have benefited from the theory and designs: flows, flexible structures, materials, thermal and chemically reacting dynamics, energy (from oil drilling to batteries and magnetic confinement fusion), and vehicles
A stability result on optimal Skorokhod embedding
No full textMotivated by the model- independent pricing of derivatives calibrated to the real market, we consider an optimization problem similar to the optimal Skorokhod embedding problem, where the embedded Brownian motion needs only to reproduce a finite number of prices of Vanilla options. We derive in this paper the corresponding dualities and the geometric characterization of optimizers. Then we show a stability result, i.e. when more and more Vanilla options are given, the optimization problem converges to an optimal Skorokhod embedding problem, which constitutes the basis of the numerical computation in practice. In addition, by means of different metrics on the space of probability measures, a convergence rate analysis is provided under suitable conditions
DS-Mamba: Dynamic snake visual state space model for vessel segmentation
No full textInternational audienceAccurately segmenting vascular networks holds significant clinical implications for disease diagnosis and analysis. However, the intrinsically elongated, serpentine, and multi-scale nature of these structures poses a significant challenge, with existing methods often struggling to preserve both global connectivity and local morphological fidelity. To address this challenge, we propose a novel deep-learning architecture, termed Dynamic Snake Mamba (DS-Mamba), inspired by the sinuous morphology of vessels. DS-Mamba first leverages a Mamba backbone, composed of Residual Visual State Space (ResVSS) blocks, to establish a topologically coherent global representation of the vascular network. Subsequently, Dynamic Snake Convolutions (DSC) are strategically embedded to enhance the feature extraction of local serpentine details. To further improve its capabilities, the architecture incorporates three key components: (1) a Multi-scale Information Mamba Fusion (MIMF) mechanism that aggregates features from all encoder stages; (2) a Snake Tokenized Kolmogorov-Arnold Network (STK) at the bottleneck to manage complex feature interactions; and (3) Global-Local Information Fusion (GLIF) blocks that merge the global context with serpentine details. The efficacy of DS-Mamba was validated through comprehensive experiments on eight diverse tubular structure datasets. Results demonstrate that our approach not only achieves state-of-the-art performance in connectivity and morphological fidelity but also exhibits superior accuracy in segmenting thin, low-contrast vessels and robust resilience against high-intensity image noise. Furthermore, rigorous capacity-controlled ablation studies confirm that the performance gains stem from the synergistic architectural design rather than parameter scaling. Finally, inference speed analysis verifies the model's feasibility for real-time clinical applications