Speedup Factor Research Articles

Machine learning predictor for ‘measurement-to-track’ association for the ATLAS Inner Detector trigger

The track finding algorithm adopted for the LHC Run-2 data taking period is based on combinatorial track following, where the number of seeds scales non-linearly with the number of hits. The corresponding CPU time increase, close to cubical, creates huge and ever-increasing demand for computing power. This is particularly problematic for the silicon tracking detectors, where the hit occupancy is the largest. Therefore, it is essential that resource use is reduced, whilst maintaining the ability to reconstruct tracks with minimal loss in efficiency. This paper briefly summarises the work that has been done to optimise the HLT ID track seeding software for ATLAS Run-3 and beyond, in order to reduce the number of fake seeds constructed. An ML-based algorithm has been developed to predict if a pair of hits belong to the same track given input hit features, focusing on cluster width and inverse track inclination. The implementation of the trained predictor in the form of Look-Up Tables is presented, along with performance results in terms of tracking efficiency and speed-up factor using simulated data.

Efficient parallel strategy for molecular plasmonics – A numerical tool for integrating Maxwell-Schrödinger equations in three dimensions

An efficient parallelization approach to simulate optical properties of ensembles of quantum emitters in realistic electromagnetic environments is considered. It relies on balancing computing load of utilized processors and is built into three-dimensional domain decomposition methodology implemented for numerical integration of Maxwell's equations. The approach employed enables directly accessing dynamics of collective effects as the number of molecules in simulations can be drastically increased. Numerical experiments measuring speedup factors demonstrate the efficiency of the proposed methodology. As an example, we consider dynamics of nearly 700,000 diatomic molecules with ro-vibrational degrees of freedom explicitly accounted for coupled to electromagnetic radiation crafted by periodic arrays of split-ring resonators and triangular nanoholes. As an application of the approach, dissociation dynamics under strong coupling conditions is scrutinized. It is demonstrated that the dissociation rates are significantly affected near polaritonic frequencies.

Variational Reduced-Order Modeling of Thermomechanical Shape Memory Alloy Based Cooperative Bistable Microactuators

This article presents the formulation and application of a reduced-order thermomechanical finite strain shape memory alloy (SMA)-based microactuator model for switching devices under thermal loading by Joule heating. The formulation is cast in the generalized standard material framework with an extension for thermomechanics. The proper orthogonal decomposition (POD) is utilized for capturing a reduced basis from a precomputed finite element method (FEM) full-scale model. The modal coefficients are computed by optimization of the underlying incremental thermomechanical potential, and the weak form for the mechanical and thermal problem is formulated in reduced-order format. The reduced-order model (ROM) is compared with the FEM model, and the exemplary mean absolute percentage errors for the displacement and temperature are 0.973% and 0.089%, respectively, with a speedup factor of 9.56 for a single SMA-based actuator. The ROM presented is tested for single and cooperative beam-like actuators. Furthermore, cross-coupling effects and the bistability phenomenon of the microactuators are investigated.

Accelerating metabolic models evaluation with statistical metamodels: application to Salmonella infection models

Mathematical and numerical models are increasingly used in microbial ecology to model the fate of microbial communities in their ecosystem. These models allow to connect in a mechanistic framework species-level informations, such as the microbial genomes, with macro-scale features, such as species spatial distributions or metabolite gradients. Numerous models are built upon species-level metabolic models that predict the metabolic behaviour of a microbe by solving an optimization problem knowing its genome and its nutritional environment. However, screening the community dynamics with these metabolic models implies to solve such an optimization problem by species at each time step, leading to a significant computational load further increased by several orders of magnitude when spatial dimensions are added. In this paper, we propose a statistical framework based on Reproducing Kernel Hilbert Space (RKHS) metamodels that are used to provide fast approximations of the original metabolic model. The metamodel can replace the optimization step in the system dynamics, providing comparable outputs at a much lower computational cost. We will first build a system dynamics model of a simplified gut microbiota composed of a unique commensal bacterial strain in interaction with the host and challenged by a Salmonella infection. Then, the machine learning method will be introduced, and particularly the ANOVA-RKHS that will be exploited to achieve variable selection and model parsimony. A training dataset will be constructed with the original system dynamics model and hyper-parameters will be carefully chosen to provide fast and accurate approximations of the original model. Finally, the accuracy of the trained metamodels will be assessed, in particular by comparing the system dynamics outputs when the original model is replaced by its metamodel. The metamodel allows an overall relative error of 4.71% but reducing the computational load by a speed-up factor higher than 45, while correctly reproducing the complex behaviour occurring during Salmonella infection. These results provide a proof-of-concept of the potentiality of machine learning methods to give fast approximations of metabolic model outputs and pave the way towards PDE-based spatio-temporal models of microbial communities including microbial metabolism and host-microbiota-pathogen interactions.

Real-Time Surface-Based Volume Constraints on Mass-Spring Model in Unity3D

This paper describes a parallel method to simulate real-time 3D deformable objects using volume preservation constraints on a mass-spring model (MSM) to achieve plausible results in real-time performance. Instead of considering a volumetric mesh which is mostly used to simulate deformable objects, we purely take the surface of the 3D object into account to reduce the time complexity and obtain a high-quality deformable. In the conventional MSM, we can simply control the shape of the deformable object through the stiffness and damping coefficients which is beneficial for our volume constraint. We use the divergence theorem and implicit constraint enforcement scheme to maintain the volume of the object and deform it freely. The surface-based volume constraint is applied to correct the force of the spring network. The proposed algorithm was designed on compute shader in Unity3D and runs outside the normal rendering pipeline in the graphics processing unit (GPU) in order to utilize the massive parallel process to accelerate the performance of the simulation. The performance of the simulation can be accelerated by using the parallel processing method on the GPU with an average speedup factor of 4.27 using the conventional mass-spring method, and an average speedup factor of 2.54 for the volume preservation constraint method. We present several scenes which demonstrate the volume-preserving deformations using the conventional mass-spring method and volume-preservation constraint method and the volume loss of deformable objects for all 3D models is compared. The volume preservation method obtains a volume loss significantly lower than the conventional MSM for all experimental tests.

AI-IoT Platform for Blind Estimation of Room Acoustic Parameters Based on Deep Neural Networks

Room acoustical parameters have been widely used to describe sound perception in indoor environments, such as concert halls, conference rooms, etc. Many of them have been standardized and often have a high computational demand. With the increasing presence of deep learning approaches in automatic monitoring systems, wireless acoustic sensor networks (WASNs) offer great potential to facilitate the estimation of such parameters. In this scenario, convolutional neural networks (CNNs) offer significant reductions in the computational requirements for in-node parameter predictions, enabling the so-called Artificial Intelligence-Internet of Things (AI-IoT). In this article, we describe the design and analysis of a CNN trained to predict simultaneously a set of common room acoustical parameters directly from speech signals, without the need for specific impulse response measurements. The results show that the proposed CNN-based prediction of room acoustical parameters and speech intelligibility achieves a relative error rate of less than a 5.5%, accompanied by a computational speedup factor close to 250 with respect to the conventional signal processing approach.

Modeling for large-scale offshore wind farm using multi-thread parallel computing

Large-scale Offshore Wind Farms (OWFs) face difficulty when carrying out microsecond-level Electro-Magnetic Transient (EMT) simulations due to the large number of Wind Turbines (WTs) and electrical nodes. This paper provides an OWF modeling method considering the electrical collection network based on the integrated equivalent modeling method of single WT, which can eliminate the internal nodes and reduce the order of admittance matrix. Meanwhile, the multi-thread parallel computing is integrated into the whole solution procedure, which achieves further simulation speedup. Moreover, simulations are performed in PSCAD/EMTDC to compare the model accuracy and time efficiency of the proposed method with the detailed model. The simulation results show that the proposed method can accurately simulate the dynamic characteristics of large-scale OWFs with large speedup factors.

Using remote GPU virtualization techniques to enhance edge computing devices

The Internet of Things (IoT) is driving the next economic revolution where the main actors are both data and immediacy. The IoT ecosystem is increasingly generating large amounts of data that are created but never analyzed. Efficient big data analysis in IoT infrastructures is becoming mandatory to transform this data deluge into meaningful information. Edge computing is proving to be a compelling alternative for enabling computing capabilities at the edge of the network. These computing capabilities could help in transforming the generated data into useful information. However, the edge computing platforms available on the market are low-power devices with limited computing horsepower. In this paper, we present a novel approach to providing computing resources to edge devices without penalizing their power consumption by using remotely virtualized GPUs. We evaluate this hardware environment by executing a computational-intensive clustering algorithm called Fuzzy Minimals (FM). Our results show that using a remotely virtualized GPU on the edge device provides a 3.2x speed-up factor compared to the local counterpart version. Moreover, we report up to 30% reduction in power consumption and up to 80% of energy savings at the edge device, delegating the GPU workload to the backend, transparently to the programmer.

Capability Extension of the High-Resolution Thermal-Hydraulic Code ESCOT for Hexagonal Geometry Core Multiphysics Analysis

The capability of the ESCOT pin-level nuclear reactor core thermal-hydraulic (T/H) code is extended for the multiphysics analysis of hexagonal geometry cores, and its performance is assessed by a code-to-code comparison with COBRA-TF (CTF). ESCOT is an accurate yet fast core T/H solution aimed at high-fidelity and high-resolution multiphysics core analysis in the framework of massively parallel computing platforms. The coupling of ESCOT with the nTRACER direct whole-core calculation code is enhanced for the hexagonal geometry handling needed for VVER core analysis. The lateral momentum terms, the turbulent mixing coefficient values, and the parallelization algorithms are modified to handle hexagonal geometry. The newly implemented ESCOT features are verified by comparing single-assembly and full-core steady-state standalone and coupled solutions for the VVER-1000 benchmark X-2 with CTF results. The ESCOT and CTF results show differences within an acceptable range in both standalone and coupled calculations. The computing time superiority due to the use of the drift flux model (DFM) of ESCOT over the CTF two-fluid model is corroborated with a speedup factor of 1.5. The use of the DFM together with the axial-radial parallelization capability of ESCOT makes ESCOT an ideal alternative to replace the simplified built-in T/H solver in nTRACER as the coupled simulation results demonstrate.

Machine Learning-Enabled Prediction of Transient Injection Map in Automotive Injectors With Uncertainty Quantification

AbstractAccurate prediction of injection profiles is a critical aspect of linking injector operation with engine performance and emissions. However, highly resolved injector simulations can take one to two weeks of wall-clock time, which is incompatible with engine design cycles with desired turnaround times of less than a day. Hence, it is important to reduce the time-to-solution of the internal flow simulations by several orders of magnitude to make it compatible with engine simulations. This work demonstrates a data-driven approach for tackling the computational overhead of injector simulations, whereby the transient injection profiles are emulated for a side-oriented, single-hole diesel injector using a Bayesian machine-learning framework. First, an interpretable Bayesian learning strategy was employed to understand the effect of design parameters on the total void fraction field. Then, autoencoders are utilized for efficient dimensionality reduction of the flowfields. Gaussian process models are finally used to predict the spatiotemporal void fraction field at the injector exit for unknown operating conditions. The Gaussian process models produce principled uncertainty estimates associated with the emulated flowfields, which provide the engine designer with valuable information of where the data-driven predictions can be trusted in the design space. The Bayesian flowfield predictions are compared with the corresponding predictions from a deep neural network, which has been transfer-learned from static needle simulations from a previous work by the authors. The emulation framework can predict the void fraction field at the exit of the orifice within a few seconds, thus achieving a speed-up factor of up to 38 × 106 over the traditional simulation-based approach of generating transient injection maps.

Higher-order dynamic mode decomposition on-the-fly: A low-order algorithm for complex fluid flows

This article presents a new method to identify the main patterns describing the flow motion in complex flows. The algorithm is an extension of the higher-order dynamic mode decomposition (HODMD), which compresses the snapshots from the analysed database and progressively updates new compressed snapshots on-the-fly, so it is denoted as HODMD on-the-fly (HODMD-of). This algorithm can be applied in parallel to the numerical simulations (or experiments), and it exhibits two main advantages over offline algorithms: (i) it automatically selects on-the-fly the number of necessary snapshots from the database to identify the relevant dynamics; and (ii) it can be used from the beginning of a numerical simulation (or experiment), since it uses a sliding-window to automatically select, also on-the-fly, the suitable interval to perform the data analysis, i.e. it automatically identifies and discards the transient dynamics. The HODMD-of algorithm is suitable to build reduced order models, which have a much lower computational cost than the original simulation. The performance of the method has been tested in three different cases: the axi-symmetric synthetic jet, the three-dimensional wake of a circular cylinder and the turbulent wake behind a wall-mounted square cylinder. The obtained speed-up factors are around 7 with respect to HODMD; this value depends on the simulation and the configuration of the hyperparameters. HODMD-of also provides a significant reduction of the memory requirements, between 40−80% amongst the two- and three-dimensional cases studied in this paper.

Space‐local reduced‐order bases for accelerating reduced‐order models through sparsity

AbstractProjection‐based model order reduction (PMOR) methods based on linear or affine approximation subspaces accelerate numerical predictions by reducing the dimensionality of the underlying computational models. The state of the art of PMOR includes approximation methods based on state‐local subspaces—that is, subspaces associated with different regions of the solution manifold—and methods based on adaptive reduced‐order bases. For challenging applications such as those associated with highly nonlinear problems, such methods accelerate traditional PMOR by controlling the dimension of the reduced‐order basis (ROB) and associated projection‐based reduced‐order model (PROM). This article proposes an alternative as well as complementary approach for accelerating PMOR based on introducing sparsity into the ROB, in order to enhance the computational efficiency of the associated PROM. Specifically, the proposed approach introduces sparsity in PMOR by partitioning the computational domain rather than, or in addition to, the solution manifold, and therefore leads to the concept of a space‐local ROB. This concept is compatible with both concepts of a state‐local ROB and hyperreduction. It is demonstrated for two computational fluid dynamics problems in turbulent flow applications. Acceleration factors of the order of 1.5 relative to traditional PMOR and CPU time speedup factors of several orders of magnitude relative to high‐dimensional models are reported.

An Efficient Model-Compressed EEGNet Accelerator for Generalized Brain-Computer Interfaces with Near Sensor Intelligence.

Brain-computer interfaces (BCIs) is promising in interacting with machines through electroencephalogram (EEG) signal. The compact end-to-end neural network model for generalized BCIs, EEGNet, has been implemented in hardware to get near sensor intelligence, but without enough efficiency. To utilize EEGNet in low-power wearable device for long-term use, this paper proposes an efficient EEGNet inference accelerator. Firstly, the EEGNet model is compressed by embedded channel selection, normalization merging, and product quantization. The customized accelerator based on the compressed model is then designed. The multilayer convolutions are achieved by reusing multiplying-accumulators and processing elements (PEs) to minimize area of logic circuits, and the weights and intermediate results are quantized to minimize memory sizes. The PEs are clock-gated to save power. Experimental results in FPGA on three datasets show the good generalizing ability of the proposed design across three BCI diagrams, which only consumes 3.31% area and 1.35% power compared to the one-to-one parallel design. The speedup factors of 1.4, 3.5, and 3.7 are achieved by embedded channel selection with negligible loss of accuracy (-0.80%). The presented accelerator is also synthesized in 65nm CMOS low power (LP) process and consumes 0.23M gates, 24.4ms/inference, 0.267mJ/inference, which is 87.22% more efficient than the implementation of EEGNet in a RISC-V MCU realized in 40nm CMOS LP process in terms of area, and 20.77% more efficient in terms of energy efficiency on BCIC-IV-2a dataset.

Communication: Weakening the critical dynamical slowing down of models with SALR interactions.

In systems with frustration, the critical slowing down of the dynamics severely impedes the numerical study of phase transitions for even the simplest of lattice models. In order to help sidestep the gelation-like sluggishness, a clearer understanding of the underlying physics is needed. Here, we first obtain generic insight into that phenomenon by studying one-dimensional and Bethe lattice versions of a schematic frustrated model, the axial next-nearest neighbor Ising (ANNNI) model. Based on these findings, we formulate two cluster algorithms that speed up the simulations of the ANNNI model on a 2D square lattice. Although these schemes do not eliminate the critical slowing own, speed-ups of factors up to 40 are achieved in some regimes.

Data-driven linear time advance operators for the acceleration of plasma physics simulation

We demonstrate the application of data-driven linear operator construction for time advance with a goal of accelerating plasma physics simulation. We apply dynamic mode decomposition (DMD) to data produced by the nonlinear SOLPS-ITER (Scrape-off Layer Plasma Simulator - International Thermonuclear Experimental Reactor) plasma boundary code suite in order to estimate a series of linear operators and monitor their predictive accuracy via online error analysis. We find that this approach defines when these dynamics can be represented by a sequence of approximate linear operators and is essential for providing consistent projections when compared to an unconstrained application. For linear diffusion and advection–diffusion fluid test problems, we construct and apply operators within explicit and implicit time advance schemes, demonstrating that stability can be robustly guaranteed in each case. We further investigate the use of the linear time advance operators within several integration methods including forward Euler, backward Euler, and the matrix exponential. The application of this method to simulation data from SOLPS-ITER, with varying levels of Markov chain Monte Carlo numerical noise, shows that constrained DMD operators yield a capability to identify, extract, and integrate a (slow) subset of the present timescales. Example applications show that for projected speedup factors of 2×, 4×, and 8×, a mean relative error of 3%, 5%, and 8% and maximum relative error less than 20% are achievable, which appears acceptable for typical SOLPS-ITER steady-state simulations.

A spatiotemporal two-level method for high-fidelity thermal analysis of laser powder bed fusion

Numerical simulation of the laser powder bed fusion (LPBF) procedure for additive manufacturing (AM) is difficult due to the presence of multiple scales in both time and space, ranging from the part scale (order of centimeters/seconds) to the powder scale (order of microns/milliseconds). This difficulty is compounded by the fact that the regions of small-scale behavior are not fixed, but change in time as the geometry is produced. While much work in recent years has been focused on resolving the problem of multiple scales in space, there has been less work done on multiscale approaches for the temporal discretization of LPBF problems. In the present contribution, we extend on a previously introduced two-level method in space by combining it with a multiscale time integration method. The unique transfer of information through the transmission conditions allows for interaction between the space and time scales while reducing computational costs. At the same time, all of the advantages of the two-level method in space (namely its geometrical flexibility and the ease in which one may deploy structured, uniform meshes) remain intact. Adopting the proposed multiscale time integration scheme, we observe a computational speed-up by a factor × 2 . 44 compared to the same two-level approach with uniform time integration, when simulating a laser source traveling on a bare plate of nickel-based superalloy material following an alternating scan path of fifty laser tracks. • A Two-level method is extended to a spatiotemporal formulation without continuous mesh update. • The two-level formulation allows for full scale separation, such that different spatial and temporal approximations are used on each scale. • An overall speed-up factor of approximately 2.5 is observed while the temperature profile closely captures reference solutions.

CosmicNet II: emulating extended cosmologies with efficient and accurate neural networks

In modern analysis pipelines, Einstein-Boltzmann Solvers (EBSs) are an invaluable tool for obtaining CMB and matter power spectra. To significantly accelerate the computation of these observables, the CosmicNet strategy is to replace the usual bottleneck of an EBS, which is the integration of a system of differential equations for linear cosmological perturbations, by trained neural networks. This strategy offers several advantages compared to the direct emulation of the final observables, including very small networks that are easy to train in high-dimensional parameter spaces, and which do not depend by construction on primordial spectrum parameters nor observation-related quantities such as selection functions. In this second CosmicNet paper, we present a more efficient set of networks that are already trained for extended cosmologies beyond ΛCDM, with massive neutrinos, extra relativistic degrees of freedom, spatial curvature, and dynamical dark energy. We publicly release a new branch of the class code, called classnet, which automatically uses networks within a region of trusted accuracy. We demonstrate the accuracy and performance of classnet by presenting several parameter inference runs from Planck, BAO and supernovae data, performed with classnet and the cobaya inference package. We have eliminated the perturbation module as a bottleneck of the EBS, with a speedup that is even more remarkable in extended cosmologies, where the usual approach would have been more expensive while the network's performance remains the same. We obtain a speedup factor of order 150 for the emulated perturbation module of class. For the whole code, this translates into an overall speedup factor of order 3 when computing CMB harmonic spectra (now dominated by the highly parallelizable and further optimizable line-of-sight integration), and of order 50 when computing matter power spectra (less than 0.1 seconds even in extended cosmologies).

Fast numerical propagation in high-NA imaging using the resampling angular spectrum method.

Numerical propagation calculation is a fundamental research topic in optical engineering. The standard angular spectrum method (ASM) is accurate but time- and memory-consuming, especially for high-NA systems. In this work, we propose a fast and simple numerical propagation method, the resampling ASM (RS-ASM). Numerical propagation can be accelerated by combining a resampling technique with interpolation methods in the angular spectrum domain of a constrained object at the focal plane. RS-ASM has three main advantages: simple implementation, faster calculation than the standard ASM, and SNR enhancement. Here we validate RS-ASM using theory, simulation and experiment. Using the "bilinear" ASM with a proper resampling factor can result in a speed-up factor of up to 20x (for a transformation from the angular spectrum to the E field) and 4x (for a transformation from E field to the angular spectrum), together with a SNR improvement of approximately 2x. For an application example of Gerchberg-Saxton phase reconstruction, the "bilinear" RS-ASM can converge 2.6x faster than the standard ASM.

A topology description function‐enhanced neural network for topology optimization

AbstractTopology optimization aims to find an economic and efficient structure with a lighter overall weight. Topology description functions (TDFs), which are an explicit level‐set approach for topology optimization, can obtain the explicit structure geometry function of the topology. However, due to the original hard thresholding in the conventional TDF, the TDF is a derivative‐free method that requires significant computational resources, which creates barriers to its widespread adoption among structural engineers. To fix this problem, a novel TDF‐enhanced neural network (TDF‐NN) is proposed, which introduces the sigmoid activation function as soft thresholding to allow the adoption of gradient‐based optimization approaches. Meanwhile, TDF‐NN uses a single‐layer neural network to model the TDF, and the weights of the TDF are converted to the weights of the TDF‐NN, which can be obtained directly by optimization instead of solving the linear equations. The performance of the proposed method is validated through the compliance minimization problems and the heat conduction problem, and it definitely can also be used for other topology optimization problems. The computational cost of the proposed method, the effect of the number of knots on TDF‐NN, diverse designs, and application to real‐life structures are also discussed. These results indicate that the proposed method is a universal topology optimization method. Numerical examples show that the proposed method has higher computational efficiency than the solid isotropic material with penalization method. It has a speedup factor of approximately two to 10 times. Moreover, multiple competitive solutions can be generated by changing the number of knots, which will allow architects to select a suitable structure for the design task.

General Approach to Asynchronous Circuits Simulation Using Synchronous FPGAs

Using field-programmable gate arrays (FPGAs) for software and hardware verification and development is a standard step in the digital application-specific integrated circuits (ASICs) design flow. However, asynchronous FPGAs are not available on the market and commercially available FPGAs provide support only for synchronous circuits. Although a lot of research effort has been undertaken in order to use synchronous FPGA to map asynchronous circuits, proposed solutions are typically lacking automation and target-specific circuit style and sometimes specific FPGA vendor. In this work, we present an automated solution for asynchronous circuits mapping onto the synchronous FPGAs. We build a synchronous model of the original asynchronous circuit based on the event-driven simulation concepts. The proposed approach supports a wide range of circuit styles, including those with various timing assumptions and complex circuitry structures incompatible with the standard synchronous flow. We avoid using vendor-specific features so that the model we generate can be implemented on any commercially available FPGA. We provide an extensive evaluation of our solution and demonstrate that our approach results in a speedup factor of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$1.3\times 10^{5}$ </tex-math></inline-formula> against an asynchronous circuit simulator, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$2.8\times 10^{4}$ </tex-math></inline-formula> against commercial digital simulators, and is 16.5 times slower than the expected performance of the original asynchronous circuit implemented as an ASIC.