A5027015251- Parallel Computing and Optimization Techniques
- Medical Image Segmentation Techniques
- Advanced Data Storage Technologies
- Numerical Methods and Algorithms
- Image and Signal Denoising Methods
- Embedded Systems Design Techniques
- Retinal Imaging and Analysis
- Distributed and Parallel Computing Systems
- Image Retrieval and Classification Techniques
- Retinal and Optic Conditions
- Retinal Diseases and Treatments
- Cerebrospinal fluid and hydrocephalus
- Particle accelerators and beam dynamics
- Functional Brain Connectivity Studies
- Advanced MRI Techniques and Applications
- Medical Imaging Techniques and Applications
- Advanced Image Fusion Techniques
- AI in cancer detection
- Particle Accelerators and Free-Electron Lasers
- Computational Physics and Python Applications
- Advanced Neuroimaging Techniques and Applications
- Image Processing Techniques and Applications
- Fetal and Pediatric Neurological Disorders
- Advanced Neural Network Applications
- Cell Image Analysis Techniques
Intel (United States)
2018-2025
GANIL
1991-2024
Laboratoire Images, signaux et systèmes intelligents
2005-2022
Intel (United Kingdom)
2020-2022
Paris-Est Sup
2005-2021
Université Paris-Est Créteil
2011-2021
Exascale (United Kingdom)
2020
Institut Lavoisier de Versailles
2015-2020
Université de Versailles Saint-Quentin-en-Yvelines
2012-2018
Université Paris-Saclay
2016
Numerical accuracy of floating point computation is a well studied topic which has not made its way to the end-user in scientific computing. Yet, it become critical issue with recent requirements for code modernization harness new highly parallel hardware and perform higher resolution computation. To democratize numerical analysis, important propose tools methodologies study large use cases reliable automatic way. In this paper, we verificarlo, an extension LLVM compiler automatically Monte...
Background. In recent years, deep learning has been increasingly applied to a vast array of ophthalmological diseases. Inherited retinal diseases (IRD) are rare genetic conditions with distinctive phenotype on fundus autofluorescence imaging (FAF). Our purpose was automatically classify different IRDs by means FAF images using algorithm. Methods. this study, patients retinitis pigmentosa (RP), Best disease (BD), Stargardt (STGD), as well healthy comparable group were used train multilayer...
Linear algebra computations can be greatly accelerated using spatial accelerators on FPGAs. As a standard building block of linear applications, BLAS covers wide range compute patterns that vary vastly in data reuse, bottleneck resources, matrix storage layouts, and types. However, existing implementations routines FPGAs are stuck the dilemma productivity performance. They either require extensive human effort or fail to leverage properties for acceleration. We introduce Lasa, framework...
This article presents Codelet Extractor and REplayer (CERE), an open-source framework for code isolation. CERE finds extracts the hotspots of application as isolated fragments code, called codelets . Codelets can be modified, compiled, run, measured independently from original application. Code isolation reduces benchmarking cost allows piecewise optimization Unlike previous approaches, isolates codes at compiler Intermediate Representation (IR) level. Therefore is language agnostic supports...
: To realize a tool for automatic segmentation and detection of lumbar disc degenerative disease fractures in MRI images using convolutional neural networks. We developed, on one hand, Picture Archiving Communication System (PACS) with DICOM viewer (Digital Imaging Communications Medicine) to extract training data implement 2 CNN networks dedicated the task other analysis simple pathologies hand.Two hundred forty-four scans T2 weighted sagittal were selected at university hospital Pasteur...
With an increase in awareness regarding a troubling lack of reproducibility analytical software tools, the degree validity scientific derivatives and their downstream results has become unclear. The nature issues may vary across domains, data sets, computational infrastructures, but numerical instabilities are thought to be core contributor. In neuroimaging, unexpected deviations have been observed when varying operating systems, implementations, or adding negligible quantities noise. field...
Fused Multiply-Add (FMA) functional units constitute a fundamental hardware component to train Deep Neural Networks (DNNs). Its silicon area grows quadratically with the mantissa bit count of computer number format, which has motivated adoption BrainFloat16 format (BF16). BF16 features 1 sign, 8 exponent and 7 explicit bits. Some approaches DNNs achieve significant performance benefits by using format. However, these must combine standard IEEE 754 Floating-Point 32-bit (FP32)...
.Stochastic rounding (SR) offers an alternative to the deterministic IEEE-754 floating-point modes. In some applications such as PDEs, ODEs, and neural networks, SR empirically improves numerical behavior convergence accurate solutions while theoretical background remains partial. Recent works by Ipsen, Zhou, Higham, Mary have computed probabilistic error bounds for basic linear algebra kernels. For example, inner product bound of forward is proportional \(\sqrt{n}u\) instead \(nu\) default...
In this study, the authors present a new image segmentation algorithm based on two-dimensional digital fractional integration (2D-DFI) that was inspired from properties of function. Although obtaining good result corresponds to finding optimal 2D-DFI order, propose alternative Legendre moments. This framework, called two dimensional and moments' (2D-DFILM), allows one include contextual information such as global object shape exploits 2D integration. The efficiency 2D-DFILM is shown by...
Exposing massive parallelism on 3D unstructured meshes computation with efficient load balancing and minimal synchronizations is challenging. Current approaches relying domain decomposition mesh coloring struggle to scale the increasing number of cores per nodes, especially new many-core processors. In this paper, we propose an hybrid approach using exploit distributed memory parallelism, Divide-and-Conquer, D&C, shared improve locality, at core level vectors. It illustrates a trade-off for...
Due to many factors such as, high transistor density, frequency, and low voltage, today's processors are more than ever subject hardware failures. These errors have various impacts depending on the location of error type processor. Because hierarchical structure compute units work scheduling, failure GPUs affect only part application. In this paper we present a new methodology characterize failures Nvidia based software micro-benchmarking platform implemented in OpenCL. We also which TESLA...
Quantifying errors and losses due to the use of Floating-Point (FP) calculations in industrial scientific computing codes is an important part Verification, Validation Uncertainty Quantification (VVUQ) process. Stochastic Arithmetic one way model estimate FP accuracy, which scales well large, codes. It exists different flavors, such as CESTAC or MCA, implemented various tools CADNA, Verificarlo Verrou. These methodologies are based on idea that accuracy can be modeled via randomness....
This study deals with information fusion for image segmentation. The evidence theory (or the Dempster–Shafer theory) allows modellisation of uncertainty and imprecision in as well combination different sources. Here, this approach is used an unsupervised framework to combine stochastic watershed segmentation which provides several results, a Hessian operator order obtain unique efficient method tested on natural images from Berkeley dataset evaluated using evaluation metrics. results surpass...
SUMMARY Characterizing performance is essential to optimize programs and architectures. The open source Adaptive Sampling Kit (ASK) measures the trade‐off in large design spaces. Exhaustively sampling all sets of parameters computationally intractable. Therefore, ASK concentrates exploration most irregular regions space through multiple adaptive strategies. paper presents architecture a set strategies, including new approach called Hierarchical Variance Sampling. ASK's usage demonstrated on...
In this paper, we propose a multivariate statistical model to characterize the inter- and intra-scale dependencies between image coefficients in oriented non-oriented sparse multiscale transforms domain. Our proposed model, namely Multivariate Bessel K Form, is based on extension of Form distribution. To establish practice, an analytical form PDF then estimate its hyperparameters. Also, compared it other models literature such as Anisotropic Generalized Gaussian Jeffrey models, order...