Intensive Computation Research Articles

SMITE: Single Molecule Imaging Toolbox Extraordinaire (MATLAB).

Fluorescence single molecule imaging comprises a variety of techniques that involve detecting individual fluorescent molecules. Many of these techniques involve localizing individual fluorescent molecules with precisions below the diffraction limit, which limits the spatial resolution of (visible) light-based microscopes. These methodologies are widely used to image biological structures at the nanometer scale by fluorescently tagging the structures of interest, elucidating details of the biological behavior observed. Two common techniques are single-molecule localization microscopy (SMLM), (Betzig et al., 2006; Fazel & Wester, 2022; Hell, 2007; Lidke et al., 2005; Rust et al., 2006; van de Linde et al., 2011) which is used to produce 2D or 3D super-resolution images of static or nearly static structures, and single-particle tracking (SPT) (Shen et al., 2017), which follows the time course of one or a very small number of moving tagged molecules. SMLM often involves distributions of particles at medium to high density, while SPT works in a very low density domain. These procedures all require intensive numerical computation, and the methods are tightly interwoven.

Mean-Field Learning for Edge Computing in Mobile Blockchain Networks

Blockchain has been leveraged to secure transactions for the m-commerce. However, the intensive computation in the mining process restricts the participation of mobile devices. Currently, some studies have deployed edge computing services to support the mining process, where edge servers managed by one Service Provider (SP) are considered. This paper investigates a more practical scenario with multiple SPs, where servers managed by different SPs have distinct capacities and prices, making miners’ offloading decisions rather complicated. To tackle the above challenges, we consider task offloading, block propagation and miner mobility comprehensively to maximize utilities of miners. Specifically, we first formulate a Markov game, and then design a learning-based offloading algorithm for off-chain computation, where a novel learning model is constructed by integrating Deep Reinforcement Learning (DRL) and Mean Field Theory (MFT) to guarantee a Nash equilibrium. Different from existing studies, each miner merely needs to respond to the average effect from others in our system, insteading of knowing policies of others. Finally, both theoritical and performance results show that our designed algorithm has superiority on average miner utilities and algorithm convergence time compared with other representative algorithms.

Computing Geographical Networks Generated by Air-Mass Movement.

As air masses move within the troposphere, they transport a multitude of components including gases and particles such as pollen and microorganisms. These movements generate atmospheric highways that connect geographic areas at distant, local, and global scales that particles can ride depending on their aerodynamic properties and their reaction to environmental conditions. In this article we present an approach and an accompanying web application called tropolink for measuring the extent to which distant locations are potentially connected by air-mass movement. This approach is based on the computation of trajectories of air masses with the HYSPLIT atmospheric transport and dispersion model, and on the computation of connection frequencies, called connectivities, in the purpose of building trajectory-based geographical networks. It is illustrated for different spatial and temporal scales with three case studies related to plant epidemiology. The web application that we designed allows the user to easily perform intensive computation and mobilize massive archived gridded meteorological data to build weighted directed networks. The analysis of such networks allowed us for example, to describe the potential of invasion of a migratory pest beyond its actual distribution. Our approach could also be used to compute geographical networks generated by air-mass movement for diverse application domains, for example, to assess long-term risk of spread from persistent or recurrent sources of pollutants, including wildfire smoke.

ETA: An Efficient Training Accelerator for DNNs Based on Hardware-Algorithm Co-Optimization.

Recently, the efficient training of deep neural networks (DNNs) on resource-constrained platforms has attracted increasing attention for protecting user privacy. However, it is still a severe challenge since the DNN training involves intensive computations and a large amount of data access. To deal with these issues, in this work, we implement an efficient training accelerator (ETA) on field-programmable gate array (FPGA) by adopting a hardware-algorithm co-optimization approach. A novel training scheme is proposed to effectively train DNNs using 8-bit precision with arbitrary batch sizes, in which a compact but powerful data format and a hardware-oriented normalization layer are introduced. Thus the computational complexity and memory accesses are significantly reduced. In the ETA, a reconfigurable processing element (PE) is designed to support various computational patterns during training while avoiding redundant calculations from nonunit-stride convolutional layers. With a flexible network-on-chip (NoC) and a hierarchical PE array, computational parallelism and data reuse can be fully exploited, and memory accesses are further reduced. In addition, a unified computing core is developed to execute auxiliary layers such as normalization and weight update (WU), which works in a time-multiplexed manner and consumes only a small amount of hardware resources. The experiments show that our training scheme achieves the state-of-the-art accuracy across multiple models, including CIFAR-VGG16, CIFAR-ResNet20, CIFAR-InceptionV3, ResNet18, and ResNet50. Evaluated on three networks (CIFAR-VGG16, CIFAR-ResNet20, and ResNet18), our ETA on Xilinx VC709 FPGA achieves 610.98, 658.64, and 811.24 GOPS in terms of throughput, respectively. Compared with the prior art, our design demonstrates a speedup of 3.65x and an energy efficiency improvement of 8.54x on CIFAR-ResNet20.

Flash-LLM: Enabling Cost-Effective and Highly-Efficient Large Generative Model Inference with Unstructured Sparsity

With the fast growth of parameter size, it becomes increasingly challenging to deploy large generative models as they typically require large GPU memory consumption and massive computation. Unstructured model pruning has been a common approach to reduce both GPU memory footprint and the overall computation while retaining good model accuracy. However, the existing solutions do not provide an efficient support for handling unstructured sparsity on modern GPUs, especially on the highly-structured tensor core hardware. Therefore, we propose Flash-LLM for enabling low-cost and highly efficient large generative model inference with the sophisticated support of unstructured sparsity on high-performance but highly restrictive tensor cores. Based on our key observation that the main bottleneck of generative model inference is the several skinny matrix multiplications for which tensor cores would be significantly under-utilized due to low computational intensity, we propose a general Load-as-Sparse and Compute-as-Dense methodology for unstructured sparse matrix multiplication (SpMM). The basic insight is to address the significant memory bandwidth bottleneck while tolerating redundant computations that are not critical for end-to-end performance on tensor cores. Based on this, we design an effective software framework for tensor core based unstructured SpMM, leveraging on-chip resources for efficient sparse data extraction and computation/memory-access overlapping. Extensive evaluations demonstrate that (1) at SpMM kernel level, Flash-LLM significantly outperforms the state-of-the-art library, i.e., Sputnik and SparTA by an average of 2.9X and 1.5X, respectively.(2) At end-to-end framework level on OPT-30B/66B/175B models, for tokens per GPU-second , Flash-LLM achieves up to 3.8X and 3.6X improvement over DeepSpeed and FasterTransformer, respectively, with significantly lower inference cost.

Optimized Discrete-Time Current Control for IPMSM Drives

Designing the current controller directly in the discrete-time domain is essential to high-speed applications, mainly due to the delay inherent to the digital control system. Active-damping (AD) is usually included to improve the disturbance rejection. However, the continuous-time form of the AD element is normally followed simply in the discrete-time domain design, which causes the dynamic performance and the anti-disturbance performance cannot be taken into account well. To easy the confliction, a novel AD loop, dedicated to the discrete-time plant, is proposed. In this design, the AD element is no longer limited to be a single real gain, as commonly seen in the continuous-time domain representing the active resistance. Alternatively, it is designed with the feedback control theory, allowing both the tracking response and disturbance response to be designed as required respectively. As a result, the dynamic response and disturbance rejection can be determined independently, and the parameter robustness can also be designed intentionally to a great extent. Besides, to reduce computation load in practice, a discrete-time IPMSM’s model with low computation intensity and high accuracy is presented in this paper. Some experimental results are reported to demonstrate the performance of the discrete-time control designed in this paper.

Ozone spectroscopy in the terahertz range from first high-resolution Synchrotron SOLEIL experiments combined with far-infrared measurements and ab initio intensity calculations

Ozone is one of the important molecules in terms of the impact on the atmospheric chemistry, climate changes, bio- and eco-systems and human health. It has a strong absorption in the microwave, terahertz and far-infrared spectral ranges where a large part of the Earth’s outgoing longwave radiation to space is located. In this work, the observations, and analyses of the ozone high-resolution spectra in the THz range recorded using the Synchrotron light source of the SOLEIL CNRS equipment are reported for the first time. Thanks to the exceptional brightness of the Synchrotron radiation and to the signal/noise ratio, it was possible to observe many more ozone transitions of the cold rotational band and the hot ν2-ν2 band in the range 0.9–6 THz compared to the previous works. In addition, we have carried out new measurements and assignments for the ν2 band. The simultaneous fit of the rotational band GS-GS, the hot band ν2-ν2 and the FIR ν2 band yielded an overall weighted standard deviation of 0.68 for 13,466 line positions within the experimental accuracy. This includes all previously available MW (with the best uncertainty 0.1 – 10 kHz), FIR data and the original SOLEIL measurements that provided experimental accuracy of 0.00005 – 0.0001 cm−1 for the best lines. Significant deviations in new experimental spectra compared to available spectroscopic databases were evidenced, particularly for the line positions and energy levels at high J, Ka rotational quantum numbers that are the most pronounced in the 4.5 – 6 THz range. Accurate ab initio calculations of line intensities combined with empirically fitted line positions were used to create new linelists that permit theoretical modelling of the transmittance in a good agreement with the Synchrotron spectra in the entire range of observations for various pressures and optical paths. The region near 100 cm−1 and above appears to be more sensitive to the temperature conditions that should be considered in atmospheric observation for the currently operational and future ground based and space missions.

Inference for the Optimum Using Linear Regression Models with Discrete Inputs

We present a multiple-comparison-with-the-best procedure to provide inference for the optimum from regression models with discrete inputs. Two applications are given to illustrate the methodology: two-level factorial designs to identify the best drug combination and order-of-addition experiments where the primary objective is to identify the sequence with the largest mean response. The methods easily accommodate restrictions limiting the inference set of conditions. We use simulation to determine the critical values. While the methods apply to any linear regression model, we identify cases that require just a single critical value, and we also show where approximations and upper bounds mitigate the need for intensive computation. We tabulate the required critical values for a variety of common applications: the main-effect model and two-factor interaction model estimated by certain two-level factorial designs, and the pairwise order model and several component-position models for estimation based on optimal order-of-addition designs. Our work greatly simplifies the problem of rigorous inference for the optimum from regression models with discrete inputs.

A Quasi-Dimensional Model of Heat Transfer between Multi-Concentric Monolith Structures

Metallic monolith structures are often used in compact reactor applications due to their superior heat transfer properties and lower pressure drop when compared to ceramic monoliths. Endothermic reactions like steam reforming depend heavily on externally supplied heat, making highly conductive supports especially useful. Simulations are invaluable for designing effective reactors with complex catalyst support structures but are conventionally resource-intensive. Additionally, few dedicated heat transfer experiments between monoliths exist in prior literature. To expand general knowledge of heat transfer between metal monolith structures, this work investigated heat exchange in concentric monoliths brazed to a common mantle. A computationally inexpensive quasi-dimensional model was developed and used to predict the heat exchange effectiveness and intrinsic heat transfer rate. The model used a discretized control volume approach and simplified geometries to reduce computational intensity. The model was calibrated against experimental data collected using a steady-state flow bench. After calibration, a parametric study was performed where monolith construction and flow conditions were varied. A parametric analysis showed that for identical catalyst space velocities and volumes, heat exchange effectiveness can be increased by 43.2% and heat transfer rates by 44.8% simply through increasing the surface area to volume ratio of the monolith. The described approach serves as an alternative framework for modeling catalytic heat exchangers without heavy computation and for quickly matching monolith geometries to their intended use and operating range.

DFT-Based Calculation of Molecular Hyperpolarizability and SFG Intensity of Symmetric and Asymmetric Stretch Modes of Alkyl Groups.

Vibrational sum frequency generation (SFG) spectroscopy has been extensively used for obtaining structural information of molecular functional groups at two-dimensional (2D) interfaces buried in the gas or liquid medium. Although the SFG experiment can be done elegantly, interpreting the measured intensity in terms of molecular orientation with respect to the lab coordinate is quite complicated. One of the main reasons is the difficulty of determining the hyperpolarizability tensors of even simple molecules that govern their SFG responses. The single-bond polarizability derivative model has been proposed to estimate the relative magnitude of SFG-active hyperpolarizability by assuming that the perturbation associated to each vibration is negligible. In this study, density functional theory was used to calculate the polarizability and dipole derivative tensors of the CH3 stretch mode of CH3I, CH3CH2I, CH3OH, and CH3CH2OH. Then, the hyperpolarizability tensors of symmetric and asymmetric vibration modes were calculated considering the Boltzmann distribution of representative conformers, which allowed us to theoretically calculate their SFG intensities at all polarization combinations as a function of the tilt angle of the CH3 group with respect to the surface normal direction. Then, the ratios of the calculated SFG intensities for the CH3 symmetric and asymmetric stretch peaks used in experimental studies for the CH3 tilt angle determination were compared. This comparison clearly showed that the effect of vibrational coupling among neighboring functional groups is significant and cannot be assumed to be negligible. This study presents new parameters that can be used in determining the average tilt angle of the CH3 group at the 2D interface with SFG measurements as well as limitations of the method.

Modelling Weather Precipitation Intensity on Surfaces in Motion with Application to Autonomous Vehicles.

With advances in the development of autonomous vehicles (AVs), more attention has been paid to the effects caused by adverse weather conditions. It is well known that the performance of self-driving vehicles is reduced when they are exposed to stressors that impair visibility or cause water or snow accumulation on sensor surfaces. This paper proposes a model to quantify weather precipitation, such as rain and snow, perceived by moving vehicles based on outdoor data. The modeling covers a wide range of parameters, such as varying the wind direction and realistic particle size distributions. The model allows the calculation of precipitation intensity on inclined surfaces of different orientations and on a circular driving path. The modeling results were partially validated against direct measurements carried out using a test vehicle. The model outputs showed a strong correlation with the experimental data for both rain and snow. Mitigation strategies for heavy precipitation on vehicles can be developed, and correlations between precipitation rate and accumulation level can be traced using the presented analytical model. A dimensional analysis of the problem highlighted the critical parameters that can help the design of future experiments. The obtained results highlight the importance of the angle of the sensing surface for the perceived precipitation level. The proposed model was used to analyze optimal orientations for minimization of the precipitation flux, which can help to determine the positioning of sensors on the surface of autonomous vehicles.

The Influence of Corotating Regions of Interaction of the Solar Wind on Long-Term Variations in the Intensity of Galactic Cosmic Rays

An analysis of the data of spacecraft that scanned large areas of the heliosphere, as well as the resultsof magnetohydrodynamic calculations, indicates that the corotating interaction regions of solar wind (SW),which are almost always present in the low- and mid-latitude heliosphere, sometimes strongly change thelarge-scale characteristics of the heliosphere that are important for long-term variations in the intensity ofgalactic cosmic rays (GCRs). In particular, for Carrington rotation no. 2066 (January–February 2008), theseregions enhance magnetic fields in the inner (r 3–5 AU) heliosphere and weaken them in the middle andfar heliosphere, as well as significantly changing the polarity distribution of heliospheric magnetic fields. Theassumption is made that in this situation the influence of the corotating interaction regions should lead to anincrease in the GCR intensity in many regions of the heliosphere. This paper discusses the process of changingthe polarity distribution of heliospheric magnetic fields due to the interaction of SW streams for Carringtonrotation no. 2066 of different speeds, the simple model of the heliospheric magnetic field without aninteraction between the SW streams of different speeds, as well as the results of numerical two-dimensionalfinite-difference calculations of longitude-averaged GCR intensity with the use of this model in comparisonwith a three-dimensional Monte Carlo calculation based on three-dimensional magnetohydrodynamic simulationof the heliosphere.

ABM-SpConv-SIMD: Accelerating Convolutional Neural Network Inference for Industrial IoT Applications on Edge Devices

Convolutional Neural Networks (CNNs) have been widely deployed, while traditional cloud data-centers based applications suffer from the bandwidth and latency network demand when applying to Industrial-Internet-of-Things (IIoT) fields. It is critical to migrate the CNN inference to edge devices for efficiency and security concerns. However, it is challenging to deploy complex CNNs on resource-constraint IIoT edge devices due to a large number of parameters and intensive floating-point computations. In this paper, we propose ABM-SpConv-SIMD, an on-device inference optimization framework, aiming at accelerating the network inference by fully utilizing the low-cost and common CPU resource. ABM-SpConv SIMD first adopts a model optimizer with pruning and quantization, which produces Sparse Convolutional models. And then, the Accumulation-Before-Multiplication mechanism is proposed to reduce multiplication operations. Additionally, the SIMD instructions, which are commonly available on cost-effective edge devices, are employed to improve the performance of convolutions. We have implemented ABM-SpConv-SIMD base on the Arm Compute Library software framework and evaluated on Hikey970 and Raspberry Pi devices with two representative models AlexNet and ResNet50. The results show that the ABM SpConv-SIMD can significantly improve the performance, and achieve on average of 1.96x and 1.73x speedup respectively over the baseline implementation with negligible loss of accuracy.

Privacy-Preserving Parallel Computation of Matrix Determinant With Edge Computing

With the widespread deployment of secure outsourcing computation, the resource-constrained client can delegate intensive computation tasks to powerful servers. Matrix determinant computation is a fundamental mathematical operation that has been widely used in IoT applications. This operation is computationally expensive. Nevertheless, the existing secure outsourcing protocols for matrix determinant are all designed based on one cloud server, which cannot well meet the low-latency and real-time computing requirements for the client. To address this issue, we explore accelerating the computation of matrix determinant by parallel outsourcing based on two nearby edge servers and propose the first practical protocol. We use the matrix blocking technique to split the computation task into multiple subtasks, which are parallel outsourced to edge servers for accelerating the computation. Moreover, we propose a privacy-preserving matrix transformation technique for data privacy protection. This technique only involves the operations of matrix-vector multiplication and matrix-matrix addition. It achieves the lightweight computation for the client and supports computational indistinguishability for the blinded input and a uniform distribution. The correctness, privacy and verifiability of the proposed protocol are analyzed. Finally, the performance advantage of the proposed protocol is demonstrated through simulation experiments.

Shaping the future of Ethereum: exploring energy consumption in Proof-of-Work and Proof-of-Stake consensus

Ethereum (ETH) is a popular Layer-1 blockchain platform that has been used to create decentralized applications (dApps) and smart contracts. Ethereum 2.0, or Serenity, is a significant update to the network that intends to address numerous issues with scalability, security, and energy efficiency. The Proof-of-Stake (PoS) consensus method will replace the Proof-of-Work (PoW) mechanism, which is one of the major new features of Ethereum 2.0. Given that PoS doesn’t require miners to do intensive mathematical calculations in order to validate transactions, it has the potential to be more energy-efficient than PoW. Additionally, this Ethereum upgrade will also be more secure due to the introduction of a new mechanism called “Casper” that will ensure that validators are always in agreement on the state of the blockchain. The paper begins by discussing the current issues facing Ethereum, including the limitations of the Proof of Work (PoW) consensus mechanism and the need for more efficient and scalable solutions. In this study, we peered at the major changes introduced by Ethereum 2.0, such as the new consensus method (Proof-of-Stake) and the addition of shard chains (Ethereum 2.0), as well as the associated development timelines, benefits and the community criticism on this upgrade.

Effective Hamiltonians for calculation of rotational energy levels and relative intensities in open-shell clusters containing O2 ([formula omitted]) and a closed–shell molecule

We introduce two effective Hamiltonians that are suitable for analysis and fitting IR and MW high resolution spectra in non-planar or planar O2-closed-shell complexes, where the closed-shell moiety is a diatom or a triatomic molecule of any symmetry. These new Hamiltonians differ in our choice for the direction of quantization of the projection of the electron-spin angular momentum. For both electron-spin coupling schemes, the total rotation-spin-tunneling Hamiltonians include tunneling, electron-spin–spin coupling, electron-spin-rotation interaction, and centrifugal distortion forces. In addition, we introduce the appropriate molecular symmetry treatment for an O2 (3Σg-)-XY2 (e.g., O2-SO2 and O2-H2O) dimer in which the monomers exhibit permeation-inversion tunneling motion. Non-vanishing matrix elements of the total Hamiltonians and expectation values of six quantum numbers are evaluated in the appropriate basis set S,Ps;P,J,MJ. Diagonalization of the total Hamiltonian matrix provides the energy levels while the eigenfunctions are used to transform expectation values of the quantum numbers into the eigenfunctions basis and for calculation of relative intensities of the allowed transitions. The reported Hamiltonians were successfully applied for fitting the observed IR and FTMW spectra of the O2-DF and O2-SO2 complexes, respectively.

Lightweight and efficient dual-path fusion network for iris segmentation

In order to tackle limitations of current iris segmentation methods based on deep learning, such as an enormous amount of parameters, intensive computation and excessive storage space, a lightweight and efficient iris segmentation network is proposed in this article. Based on the classical semantic segmentation network U-net, the proposed approach designs a dual-path fusion network model to integrate deep semantic information and rich shallow context information at multiple levels. Our model uses the depth-wise separable convolution for feature extraction and introduces a novel attention mechanism, which strengthens the capability of extracting significant features as well as the segmentation capability of the network. Experiments on four public datasets reveal that the proposed approach can raise the MIoU and F1 scores by 15% and 9% on average compared with traditional methods, respectively, and 1.5% and 2.5% on average compared with the classical semantic segmentation method U-net and other relevant methods. Compared with the U-net, the proposed approach reduces about 80%, 90% and 99% in terms of computation, parameters and storage, respectively, and the average run time up to 0.02 s. Our approach not only exhibits a good performance, but also is simpler in terms of computation, parameters and storage compared with existing classical semantic segmentation methods.

The system for distributed energy resources testing according to the IEEE 1547-2018 standard

PurposeThis paper aims to propose a new approach to testing distributed energy resources (DERs) in compliance with the IEEE 1547-2018 standard and describes a new, integrated testing and validation system.Design/methodology/approachThe system is built on the virtual instrumentation paradigm, using acquisition modules to measure physical quantities, while signal processing, including intensive calculations of required parameters, data processing, manipulation and reporting are performed on a computing device.FindingsIntensive laboratory measurements were performed on a laboratory prototype of a microgrid that emulates DERs. The results obtained using the system described were compared with the measurements obtained by the reference instruments. As all the results match, the usability of the system was verified.Practical implicationsThis approach to the realization of the testing and validation system has obvious advantages compared to the classical instruments and provides significant flexibility in multiple aspects. First, the system described integrates all the functions of different instruments into one measuring system, making the entire testing and validation process significantly cheaper and faster. Second, the implementation of the system is possible on different computing platforms depending on specific needs. Third, the software implementation of the system functions enables simple upgrading and the introduction of new functions or changes to existing ones according to changes in the standard. Finally, the system described is designed to automatically provide reports on compliance with the standard.Originality/valueThis paper emphasizes the advantages of the proposed approach over classical testing. The value of the paper is reflected in the applicability and practical implications of the proposed and described hardware and software technical solutions.

Topology optimization of geometrically nonlinear structures using reduced-order modeling

High computational costs are encountered in topology optimization problems of geometrically nonlinear structures since intensive use has to be made of incremental-iterative finite element simulations. To alleviate this computational intensity, reduced-order models (ROMs) are explored in this paper. The proposed method targets ROM bases consisting of a relatively small set of base vectors while accuracy is still guaranteed. For this, several fully automated update and maintenance techniques for the ROM basis are investigated and combined. In order to remain effective for flexible structures, path derivatives are added to the ROM basis. The corresponding sensitivity analysis (SA) strategies are presented and the accuracy and efficiency are examined. Various geometrically nonlinear examples involving both solid as well as shell elements are studied to test the proposed ROM techniques. Test cases demonstrates that the set of degrees of freedom appearing in the nonlinear equilibrium equation typically reduces to several tenth. Test cases show a reduction of up to 6 times fewer full system updates.

Combining kernel principal component analysis and spatial group-wise enhance convolutional neural network for fault recognition of rolling element bearings

Data-driven deep learning methods have been widely used in the fault diagnosis of rolling bearings, while general network structures are complex with numerous parameters and computationally intensive calculations, leading to limited real-time performance and delayed fault detection. To address these challenges, this paper presents a novel hybrid framework, termed FKP-SGECNN, for efficient and accurate bearing fault identification. The proposed framework combines the strengths of kernel principal component analysis (KPCA), Fisher criterion, spatial group-wise enhance network (SGENet), and convolutional neural network. In the proposed framework, FKP incorporates Fisher criterion to optimize the kernel functions in KPCA, effectively reducing information redundancy in the input data. Furthermore, SGENet is integrated to streamline the network structure and enhance the model’s generalization capability, while maintaining high diagnostic accuracy. The performance of the hybrid framework implies a great potential, which was evaluated by several case studies using multi-class data of bearing faults.