Parallel Framework Research Articles

DAMNet: Dynamic mobile architectures for Alzheimer's disease.

Alzheimer's disease (AD) presents a significant challenge in healthcare, highlighting the necessity for early and precise diagnostic tools. Our model, DAMNet, processes multi-dimensional AD data effectively, utilizing only 7.4 million parameters to achieve diagnostic accuracies of 98.3% in validation and 99.9% in testing phases. Despite a 20% pruning rate, DAMNet maintains consistent performance with less than 0.2% loss in accuracy. The model also excels in handling 3D (Three-Dimensional) MRI data, achieving a 95.7% F1 score within 805s during a rigorous three-fold validation over 200 epochs. Furthermore, we introduce a novel parallel intelligent framework for early AD detection that improves feature extraction and incorporates advanced data management and control. This framework sets a new benchmark in intelligent, precise medical diagnostics, adeptly managing both 2D (Two-Dimensional) and 3D imaging data.

Parallel Primal-Dual Method with Linearization for Structured Convex Optimization

This paper presents the Parallel Primal-Dual (PPD3) algorithm, an innovative approach to solving optimization problems characterized by the minimization of the sum of three convex functions, including a Lipschitz continuous term. The proposed algorithm operates in a parallel framework, simultaneously updating primal and dual variables, and offers potential computational advantages. This parallelization can greatly accelerate computation, particularly when run on parallel computing platforms. By departing from traditional primal-dual methods that necessitate strict parameter constraints, the PPD3 algorithm removes reliance on the spectral norm of the linear operator, significantly reducing the computational burden associated with its evaluation. As the problem size grows, calculating the spectral norm, which is essential for many primal-dual methods, becomes progressively more expensive. In addition, adaptive step sizes are computed to accelerate the convergence process. In contrast to most primal-dual approaches that employ a fixed step size constrained by a global upper limit throughout all iterations, the adaptive step size is typically greater and may result in faster convergence. An O(1/k) ergodic convergence rate is proved theoretically. Applications in Fused LASSO and image inpainting demonstrate the method’s efficiency in computation time and convergence rate compared to state-of-the-art algorithms.

Coupling immersed boundary method with wall function for high Reynolds number compressible flows on adaptive Cartesian grids

In this paper, a high Reynolds number compressible flow simulation method based on the immersed boundary method (IBM) is developed on the adaptive Cartesian grids to address two problems: one is the ability of large-scale Cartesian grid generation, and the other is the resolution of thin shear layers. For the former, an efficient automatic parallel generation method of adaptive Cartesian grids is proposed based on the k-d tree theory. The method has good parallel scalability and computational efficiency, e.g., the generation of 1.41 × 109 cells by 2048 cores takes only 0.40 min. For the latter, an IBM-wall function coupling method is designed to simulate geometrical configurations with high Reynolds numbers in the parallel framework. Numerical experiments show that the presented method can yield results consistent with the use of body-fitted grids. In conclusion, the presented method can realize a fully automatic simulation of high Reynolds number compressible flows, significantly reducing the labor cost in the computational fluid dynamics process, and can improve the accuracy and efficiency of engineering applications.

A Parallel Framework for Fast Charge/Discharge Scheduling of Battery Storage Systems in Microgrids

Fast charge/discharge scheduling of battery storage systems is essential in microgrids to effectively balance variable renewable energy sources, meet fluctuating demand, and maintain grid stability. To achieve this, parallel processing is employed, allowing batteries to respond instantly to dynamic conditions. By managing the complexity, high data volume, and rapid decision-making requirements in real time, parallel processing ensures that the microgrid operates with stability, efficiency, and safety. With the application of deep reinforcement learning (DRL) in scheduling algorithm design, the demand for computational power has further increased significantly. To address this challenge, we propose a Ray-based parallel framework to accelerate the development of fast charge/discharge scheduling for battery storage systems in microgrids. We demonstrate how to implement a real-world scheduling problem in the framework. We focused on minimizing power losses and reducing the ramping rate of net loads by leveraging the Asynchronous Advantage Actor Critic (A3C) algorithms and the features of the Ray cluster for real-time decision making. Multiple instances of OpenDSS were executed concurrently, with each instance simulating a distinct environment and efficiently processing input data. Additionally, Numba CUDA was utilized to facilitate GPU acceleration of shared memory, significantly enhancing the performance of the computationally intensive reward function in A3C. The proposed framework enhanced scheduling performance, enabling efficient energy management in complex, dynamic microgrid environments.

DNEA: an R package for fast and versatile data-driven network analysis of metabolomics data

BackgroundMetabolomics is a high-throughput technology that measures small molecule metabolites in cells, tissues or biofluids. Analysis of metabolomics data is a multi-step process that involves data processing, quality control and normalization, followed by statistical and bioinformatics analysis. The latter step often involves pathway analysis to aid biological interpretation of the data. This approach is limited to endogenous metabolites that can be readily mapped to metabolic pathways. An alternative to pathway analysis that can be used for any classes of metabolites, including unknown compounds that are ubiquitous in untargeted metabolomics data, involves defining metabolite-metabolite interactions using experimental data. Our group has developed several network-based methods that use partial correlations of experimentally determined metabolite measurements. These were implemented in CorrelationCalculator and Filigree, two software tools for the analysis of metabolomics data we developed previously. The latter tool implements the Differential Network Enrichment Analysis (DNEA) algorithm. This analysis is useful for building differential networks from metabolomics data containing two experimental groups and identifying differentially enriched metabolic modules. While Filigree is a user-friendly tool, it has certain limitations when used for the analysis of large-scale metabolomics datasets.ResultsWe developed the DNEA R package for the data-driven network analysis of metabolomics data. We present the DNEA workflow and functionality, algorithm enhancements implemented with respect to the package’s predecessor, Filigree, and discuss best practices for analyses. We tested the performance of the DNEA R package and illustrated its features using publicly available metabolomics data from the environmental determinants of diabetes in the young. To our knowledge, this package is the only publicly available tool designed for the construction of biological networks and subsequent enrichment testing for datasets containing exogenous, secondary, and unknown compounds. This greatly expands the scope of traditional enrichment analysis tools that can be used to analyze a relatively small set of well-annotated metabolites.ConclusionsThe DNEA R package is a more flexible and powerful implementation of our previously published software tool, Filigree. The modular structure of the package, along with the parallel processing framework built into the most computationally extensive steps of the algorithm, make it a powerful tool for the analysis of large and complex metabolomics datasets.

Data imbalanced fault diagnosis of gearbox transmission system under various speeds based on dynamic dual-scale normalized fusion network

Abstract Gearboxes play a pivotal role in industrial production, and their reliability and safety are essential for production safety and efficiency. However, gearboxes frequently encounter challenges such as variable rotational speeds and unknown operating conditions. Unfortunately, most existing traditional fault diagnosis methods face the following issues: (1) They heavily rely on expert experience and pre-existing knowledge bases, making them unable to tackle fault diagnosis in unknown working conditions. (2) While addressing various speed issues, they seldom consider the problem of data imbalance in real-world industrial environments. (3) Many transfer learning methods primarily focus on global distribution alignment and knowledge transfer between source and target domains, neglecting the importance of fine-grained distribution alignment between subdomains. To address these issues, a dynamic dual-scale normalization fusion network (DSNFNet) is proposed for cross-domain fault diagnosis under variable speed and data imbalance. Firstly, the two parallel graph convolution frameworks constructed are used to extract multi-scale fault features. Subsequently, a dual-scale normalization fusion module is adopted to integrate the global and local fault feature information within the subdomains of both the source and target domains, thereby aligning their fine-grained distributions to obtain domain-invariant features. Finally, a dynamic soft threshold feedback strategy is introduced, which assigns pseudo labels to the target domain data, enabling the model to give equal attention to each class of data samples, even under data imbalance conditions, thereby improving the recognition accuracy of minority fault classes. Validating the proposed method on two real cases, our method achieved the highest accuracy compared to other advanced approaches.

On a Stable Multiplicative Calculus-Based Hybrid Parallel Scheme for Nonlinear Equations

Fractional-order nonlinear equation-solving methods are crucial in engineering, where complex system modeling requires great precision and accuracy. Engineers may design more reliable mechanisms, enhance performance, and develop more accurate predictions regarding outcomes across a range of applications where these problems are effectively addressed. This research introduces a novel hybrid multiplicative calculus-based parallel method for solving complex nonlinear models in engineering. To speed up the method’s rate of convergence, we utilize a second-order multiplicative root-finding approach as a corrector in the parallel framework. Using rigorous theoretical analysis, we illustrate how the hybrid parallel technique based on multiplicative calculus achieves a remarkable convergence order of 12, indicating its effectiveness and efficiency in solving complex nonlinear equations. The intrinsic stability and consistency of the approach—when applied to nonlinear situations—are clearly indicated by the symmetry seen in the dynamical planes for various parameter values. The method’s symmetrical behavior indicates that it produces accurate findings under a range of scenarios. Using a dynamical system procedure, the ideal parameter values are systematically analyzed in order to further improve the method’s performance. Implementing the aforementioned parameter values using the parallel approach yields very reliable and consistent outcomes. The method’s effectiveness, reliability, and consistency are evaluated through the analysis of numerous nonlinear engineering problems. The analysis provides a detailed comparison with current techniques, emphasizing the benefits and potential improvements of the novel approach.

Robust deadline-aware network function parallelization framework under demand uncertainty

The orchestration of Service Function Chains (SFCs) in Mobile Edge Computing (MEC) becomes crucial for ensuring efficient service provision, especially under dynamic and uncertain demand. Meanwhile, the parallelization of Virtual Network Functions (VNFs) within an SFC can further optimize resource usage and reduce the risk of deadline violations. However, most existing works formulate the SFC orchestration problem in MEC with deterministic demands and costly runtime resource reprovisioning to handle dynamic demands. This paper introduces a Robust Deadline-aware network function Parallelization framework under Demand Uncertainty (RDPDU) designed to address the challenges posed by unpredictable fluctuations in user demand and resource availability within MEC networks. RDPDU to consider end-to-end latency for SFC assembly by modeling load-dependent processing latency and load-independent propagation latency. Also, RDPDU formulates the problem assuming uncertain demand by Quadratic Integer Programming (QIP) to be resistant to dynamic service demand fluctuations. By discovering dependencies between VNFs, the RDPDU effectively assembles multiple sub-SFCs instead of the original SFC. Finally, our framework uses Deep Reinforcement Learning (DRL) to assemble sub-SFCs with guaranteed latency and deadline. By integrating DRL into the SFC orchestration problem, the framework adapts to changing network conditions and demand patterns, improving the overall system's flexibility and robustness. Experimental evaluations show that the proposed framework can effectively deal with demand fluctuations, latency, deadline, and scalability and improve performance against recent algorithms.

High-order discontinuous Galerkin method with immersed boundary treatment for compressible flows on parallel adaptive Cartesian grids

Adaptive mesh refinement (AMR) technology and high-order methods are important means to improve the quality of simulation results and have been hotspots in the computational fluid dynamics community. In this paper, high-order discontinuous Galerkin (DG) and direct DG (DDG) finite element methods are developed based on a parallel adaptive Cartesian grid to simulate compressible flow. On the one hand, a high-order multi-resolution weighted essentially nonoscillatory limiter is proposed for DG and DDG methods. This limiter can enhance the stability of DG/DDG methods for compressible flows dominated by shock waves. It is also compact, making it suitable for the implementation of AMR with frequent refinement/coarsening. On the other hand, a coupling method of DG and immersed boundary method is proposed to simulate flow around objects. Due to the compactness of DG, the physical quantities of image points can be directly obtained through the DG/DDG polynomial of the corresponding cells. It avoids the wide interpolation stencil of traditional IBM and makes it more suitable for the parallel adaptive Cartesian grid framework in this paper. Finally, the performance of the proposed method is verified through typical two- and three-dimensional cases. The results indicate that the method proposed in this paper has low numerical dissipation in smooth areas and can effectively handle compressible flow dominated by discontinuities. Moreover, for transonic flow over a sphere, the error of results between the proposed method and direct numerical simulation is within 1%, fully validating the accuracy of the method presented in this paper.

Complex weight distinctions in Harmonic Serialism

Abstract A number of the quantity-sensitive stress systems which are characterised by complex weight distinctions have often proven challenging for parallel constraint-based metrical frameworks. A number of parallel-OT accounts are proposed to explain stress systems with multi-level weight hierarchies, in which syllables with long vowels (CVV) are more prominent than closed syllables (CVC), and closed syllables are more prominent than short open syllables (CV). The majority of these accounts are either inadequate or implausible. This paper seeks to provide an alternative account of stress systems with complex weight distinctions. Gradual prosodification, as an application of Harmonic Serialism, is the proposed framework for analysis. Maintaining gradualness and harmonic improvement in prosodic structure across the different steps of the derivation is deemed to be sufficient to achieve the required effects. The realisation of the various prominence relations illustrated in those multi-level weight hierarchies is considered to be an epiphenomenon of the sequential application of independently motivated prosodification processes, such as coda consonant moraification and foot construction. The proposed gradual prosodification model is applied to the stress systems of Kashmiri and Pulaar, that have identical algorithms for primary stress assignment, in addition to the secondary stress patten attested in Pulaar.

A parallel recursive framework for modelling time series

Abstract Time series modelling is of significance to several scientific fields. Several approaches based on statistics, machine learning or combinations have been utilized. In order to model and forecast time series a novel parallel framework based on recursive pseudoinverse matrices is proposed. This framework enables the design of arbitrary statistical and machine learning models, adaptively, from a set of potential basis functions. This unification enables compact definition of existing and new models as well as easy implementation for new massively parallel architectures. The choice of appropriate basis functions is analysed and the fitting accuracy, termination criteria and model update operations are presented. A block variant for multivariate time series is also proposed. Parallel GPU implementation and performance optimization of the framework are provided, based on mixed precision arithmetic and matrix operations. The use of different basis functions is showcased with respect to various model univariate and multivariate time series for applications such as regression, frequency estimation and automatic trend detection. Discussions on limitations and future directions of research are also provided.

Exploratory parallel hybrid sampling framework for imbalanced data classification

Current engineering application scenarios often face the challenge of imbalanced data, hybrid sampling is an effective method to deal with the imbalanced data classification issue, which can avoid the issues of overfitting and mistakenly deleting useful majority samples when using oversampling approach and undersampling approach alone. However, at present most of the hybrid sampling approaches are implemented serially, and the implementation of oversampling and undersampling approaches alone will cause mutual interference and influence between them. This study proposes a parallel hybrid sampling framework based on the idea of parallel engineering and theoretically analyzes its superiority. The experimental results show that when applied to five classification algorithms with three performance evaluation metrics,the proposed framework outperforms the two mainstream hybrid sampling frameworks. Moreover, the proposed framework can effectively reduce the time consumption of hybrid sampling process.

Large eddy simulation of highly underexpanded sonic jets from elliptical nozzles

The injection of high-pressure jets into quiescent air poses significant challenges in fluid dynamics, pertinent to various industrial engineering applications. This study used large eddy simulations on a massively parallel computational framework, employing a grid of over 600 million nodes, to investigate the behavior of highly underexpanded sonic jets from elliptical nozzles at a nozzle pressure ratio of 15. Three elliptical nozzles, with aspect ratios of 1.5, 2.2, and 3.0, each having a sectional area equivalent to that of a circular jet with a diameter of D=1 mm, were analyzed. The aim was to clarify the gasdynamic and mixing characteristics of these jets to guide the design of next-generation injectors. A detailed analysis of the flow provided insights into the mechanisms of turbulence generation and Reynolds stress anisotropy. This was achieved using the componentality contour approach and a modified barycentric color mapping scheme, offering valuable reference data for developing lower-order models. The results indicate a non-axisymmetric radial expansion of the jet boundary in all elliptical injectors, leading to an axis switch phenomenon. The use of elliptical orifices was found to reduce jet penetration, mitigating issues such as fuel impingement in small engine combustion chambers and promoting improved air–fuel mixing quality.

A zero-shot quantitative evaluation model for subsurface defects size based on ultrasonic nondestructive testing

Existing defect quantitative evaluation models requires collecting and labeling all types of defect data for training. However, such models have a poor ability for new defect, and collecting and labeling data is a costly task. A zero-shot evaluation model is proposed to solve the questions. First, a Wavelet Packet Transform-based method is developed to convert defect ultrasonic signals into time–frequency images. Second, we define defect semantics from length, width, and height to describe defects of different sizes. Third, the ResNet with embedded Brownian Distance Covariance matrix and multi-dimension parallel framework is designed to extract the image features. Then, an autoencoder is trained to embed the defect semantic into the feature space. Finally, the dimension of defect is evaluated by measuring the similarity between the features and the category centroids. Experimental results reveal that our model has better ability to evaluate the dimensions of new defect compared to others models.

A parallel coupling framework for DEM-MBD: Model verification and application

Bulk material handling equipment can be numerically studied using the DEM-MBD method, where the particulate dynamics are solved by DEM (Discrete Element Method) and the motion characteristics of the mechanical components are modeled by MBD (Multi-Body Dynamics). In this work, a two-way coupling framework is proposed for the co-simulation of our self-developed DEM solver (DEMSLab) with the commercial MBD software (RecurDyn) using a parallel staggered coupling approach. To verify the proposed coupling method, the simulation result of a torsional spring damper system is compared with the corresponding analytical solution. After that, to assess the feasibility of this method in industrial-scale applications, a series of simulations involving a wheel loader is carried out. Different simulation conditions are adopted to reveal their impact on computational efficiency and simulation accuracy. These conditions include the mesh representation of the equipment geometry, the MBD time step size, and the coupling interval. An improved mesh simplification algorithm is proposed and used to reasonably simplify the surface mesh of the wheel loader, thus lowering the computational intensity of particle-wall interactions in DEM. The comparison results indicate that the computational time can be largely saved while the simulation accuracy remains considerably high. The MBD step size and coupling interval are crucial for numerical stability and should be selected carefully.

A parallel solver framework for fully implicit monolithic fluid-structure interaction

A parallel solver framework for fully implicit monolithic fluid-structure interaction

Modeling two-phase flows with complicated interface evolution using parallel physics-informed neural networks

The physics-informed neural networks (PINNs) have shown great potential in solving a variety of high-dimensional partial differential equations (PDEs), but the complexity of a realistic problem still restricts the practical application of the PINNs for solving most complicated PDEs. In this paper, we propose a parallel framework for PINNs that is capable of modeling two-phase flows with complicated interface evolution. The proposed framework divides the problem into several simplified subproblems and solves them through training several PINNs on corresponding subdomains simultaneously. To enhance the accuracy of the parallel training framework in two-phase flow, the overlapping domain decomposition method is adopted. The optimal subnetwork sizes and partitioned method are systematically discussed, and a series of cases including a bubble rising, droplet splashing, and the Rayleigh–Taylor instability are applied for quantitative validation. The maximum relative error of quantitative values in these cases is 0.1319. Our results show that the proposed framework not only can accelerate the training procedure of PINNs, but also can capture the spatiotemporal evolution of the interface between various phases. This framework overcomes the difficulties of training PINNs to solve a forward problem in two-phase flow, and it is expected to model more realistic dynamic systems in nature.

GR-PSN: Learning to Estimate Surface Normal and Reconstruct Photometric Stereo Images.

In this paper, we propose a novel method, namely GR-PSN, which learns surface normals from photometric stereo images and generates the photometric images under distant illumination from different lighting directions and surface materials. The framework is composed of two subnetworks, named GeometryNet and ReconstructNet, which are cascaded to perform shape reconstruction and image rendering in an end-to-end manner. ReconstructNet introduces additional supervision for surface-normal recovery, forming a closed-loop structure with GeometryNet. We also encode lighting and surface reflectance in ReconstructNet, to achieve arbitrary rendering. In training, we set up a parallel framework to simultaneously learn two arbitrary materials for an object, providing an additional transform loss. Therefore, our method is trained based on the supervision by three different loss functions, namely the surface-normal loss, reconstruction loss, and transform loss. We alternately input the predicted surface-normal map and the ground-truth into ReconstructNet, to achieve stable training for ReconstructNet. Experiments show that our method can accurately recover the surface normals of an object with an arbitrary number of inputs, and can re-render images of the object with arbitrary surface materials. Extensive experimental results show that our proposed method outperforms those methods based on a single surface recovery network and shows realistic rendering results on 100 different materials.

Full QCD with milder topological freezing

We simulate Nf = 2 + 1 QCD at the physical point combining open and periodic boundary conditions in a parallel tempering framework, following the original proposal by M. Hasenbusch for 2d CPN−1 models, which has been recently implemented and widely employed in 4d SU(N) pure Yang-Mills theories too. We show that using this algorithm it is possible to achieve a sizable reduction of the auto-correlation time of the topological charge in dynamical fermions simulations both at zero and finite temperature, allowing to avoid topology freezing down to lattice spacings as fine as a ∼ 0.02 fm. Therefore, this implementation of the Parallel Tempering on Boundary Conditions algorithm has the potential to substantially push forward the investigation of the QCD vacuum properties by means of lattice simulations.

Three-dimensional computer holography based on two hybrid constraint iterative angular spectrum algorithms

A method for three-dimensional holographic image calculation based on angular spectrum layering is presented, which significantly reduces the noise resulting from image overlap during hologram reconstruction. In order to enhance the reconstruction quality, we proposed a two-hybrid constraint algorithm, which employs two parallel processing frameworks to allocate multiple plane images effectively. Additionally, by applying two different hybrid constraint factors, spatial resource allocation is utilized efficiently to mitigate the impact of axial inter-layer crosstalk noise. The numerical simulations and optical experiment results demonstrated that our method has successfully improved the reconstruction quality of holograms and proved its effectiveness and feasibility.