Search NASA⌕ Search

SEARCH · Search NASA

Results for “hardware accelerators”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 235 records · Page 13

Contrasting advantages of learning with random weights and backpropagation in non-volatile memory neural networks

Recently, a Cambrian explosion of novel, non-volatile memory (NVM) devices known as memristive devices have inspired effort in building hardware neural networks that learn like the brain. Early experimental prototypes built simple perceptrons from nanosynapses, and recently, fully-connected multi-layer perceptron (MLP) learning systems have been realized. However, while backpropagating learning systems pair well with high-precision computer memories and achieve state-of-the-art performances, this typically comes with a massive energy budget. For future Internet of Things/peripheral use cases, system energy footprint will be a major constraint, and emerging NVM devices may fill the gap by sacrificing high bit precision for lower energy. In this work, we contrast the well known MLP approach with the Extreme Learning Machine (ELM) or NoProp approach, which uses a large layer of random weights to improve the separability of high-dimensional tasks, and is usually considered inferior in a software context. However, we find that when taking device non-linearity into account, NoProp manages to equal hardware MLP system in terms of accuracy. While also using a sign-based adaptation of the delta rule for energy-savings, we find that NoProp can learn effectively with four to six ’bits’ of device analog capacity, while MLP requires eight bit capacity with the same rule. This may allow the requirements for memristive devices to be relaxed in the context of online learning. By comparing the energy footprint of these systems for several candidate nanosynapses, as well as realistic peripherals, we confirm that memristive NoProp systems save energy compared to MLP systems. Lastly, we show that ELM/NoProp systems can achieve better generalization abilities than nanosynaptic MLP systems when paired with pre-processing layers (which do not require backpropagated error). Collectively, these advantages make such systems worthy of consideration in future accelerators or embedded hardware.

97 MATHEMATICS AND COMPUTING↗

Practical Implementation of GPU-based Computing at the Grid Edge for Resilience Scenarios

This paper presents a practical implementation of GPU-accelerated computing at the grid edge to enhance power system resilience through next-generation smart meters. Advanced Metering Infrastructure (AMI) systems rely predominantly on centralized processing architectures, which limit real-time response capabilities during grid disturbances. This work proposes the integration of GPU-enabled computational platforms directly within smart meter to enable local execution support for power system analytics, fault detection algorithms, and optimization routines. The proposed framework uses the Julia programming language to leverage highperformance parallel computing capabilities while maintaining code portability and development efficiency. We use two experimental scenarios to benchmark the computational feasibility of this approach: sparse linear system solutions representative of power flow analyses, and multi-stage production cost simulations incorporating unit commitment and economic dispatch operations. Results demonstrate that computationally intensive power system algorithms, such as those supporting resilience scenario calculations, can be effectively executed at the distribution edge using commercially available embedded GPU hardware. Keywords—GPU acceleration, edge computing, smart meters, grid resilience, AMI, resilience.

De Souza, Reubun [School of Electrical Engineering↗

A Length Adaptive Algorithm-Hardware Co-design of Transformer on FPGA Through Sparse Attention and Dynamic Pipelining

Transformers are considered one of the most important deep learning models since 2018, in part because it establishes state-of-the-art (SOTA) records and could potentially replace existing Deep Neural Networks (DNNs). Despite the remarkable triumphs, the prolonged turnaround time of Transformer models is a widely recognized roadblock. The variety of sequence lengths imposes additional computing overhead where inputs need to be zero-padded to the maximum sentence length in the batch to accommodate the parallel computing platforms. This paper targets the field-programmable gate array (FPGA) and proposes a coherent sequence length adaptive algorithm–hardware co-design for Transformer acceleration. Particularly, we develop a hardware-friendly sparse attention operator and a length-aware hardware resource scheduling algorithm. The proposed sparse attention operator brings the complexity of attention-based models down to linear complexity and alleviates the off-chip memory traffic. The proposed length-aware resource hardware scheduling algorithm dynamically allocates the hardware resources to fill up the pipeline slots and eliminates bubbles for NLP tasks. Experiments show that our design has very small accuracy loss and has 80.2 × and 2.6 × speedup compared to CPU and GPU implementation, and 4 × higher energy efficiency than state-of-the-art GPU accelerator optimized via CUBLAS GEMM.

Peng, Hongwu↗

Four-dimensional volume-of-interest reconstruction for cone-beam computed tomography-guided radiation therapy

Purpose: Data sufficiency are a major problem in four-dimensional cone-beam computed tomography (4D-CBCT) on linear accelerator-integrated scanners for image-guided radiotherapy. Scan times must be in the range of 4-6 min to avoid undersampling artifacts. Various image reconstruction algorithms have been proposed to accommodate undersampled data acquisitions, but these algorithms are computationally expensive, may require long reconstruction times, and may require algorithm parameters to be optimized. The authors present a novel reconstruction method, 4D volume-of-interest (4D-VOI) reconstruction which suppresses undersampling artifacts and resolves lung tumor motion for undersampled 1-min scans. The 4D-VOI reconstruction is much less computationally expensive than other 4D-CBCT algorithms. Methods: The 4D-VOI method uses respiration-correlated projection data to reconstruct a four-dimensional (4D) image inside a VOI containing the moving tumor, and uncorrelated projection data to reconstruct a three-dimensional (3D) image outside the VOI. Anatomical motion is resolved inside the VOI and blurred outside the VOI. The authors acquired a 1-min. scan of an anthropomorphic chest phantom containing a moving water-filled sphere. The authors also used previously acquired 1-min scans for two lung cancer patients who had received CBCT-guided radiation therapy. The same raw data were used to test and compare the 4D-VOI reconstruction with the standard 4D reconstruction and the McKinnon-Bates (MB) reconstruction algorithms. Results: Both the 4D-VOI and the MB reconstructions suppress nearly all the streak artifacts compared with the standard 4D reconstruction, but the 4D-VOI has 3-8 times greater contrast-to-noise ratio than the MB reconstruction. In the dynamic chest phantom study, the 4D-VOI and the standard 4D reconstructions both resolved a moving sphere with an 18 mm displacement. The 4D-VOI reconstruction shows a motion blur of only 3 mm, whereas the MB reconstruction shows a motion blur of 13 mm. With graphics processing unit hardware used to accelerate computations, the 4D-VOI reconstruction required a 40-s reconstruction time. Conclusions: 4D-VOI reconstruction effectively reduces undersampling artifacts and resolves lung tumor motion in 4D-CBCT. The 4D-VOI reconstruction is computationally inexpensive compared with more sophisticated iterative algorithms. Compared with these algorithms, our 4D-VOI reconstruction is an attractive alternative in 4D-CBCT for reconstructing target motion without generating numerous streak artifacts.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Steady-State Calibration of a Diesel Engine in Computational Fluid Dynamics Using a Graphical Processing Unit-Based Chemistry Solver

The prospect of analysis-driven precalibration of a modern diesel engine is extremely valuable in order to significantly reduce hardware investments and accelerate engine designs compliant with stricter fuel economy regulations. Advanced modeling tools, such as CFD, are often used with the goal of streamlining significant portions of the calibration process. The success of the methodology largely relies on the accuracy of analytical predictions, especially engine-out emissions. However, the effectiveness of CFD simulation tools for in-cylinder engine combustion is often compromised by the complexity, accuracy, and computational overhead of detailed chemical kinetics necessary for combustion calculations. The standard approach has been to use skeletal kinetic mechanisms (~50 species), which consume acceptable computational time but with degraded accuracy. In this work, a comprehensive demonstration and validation of the analytical precalibration process is presented for a passenger car diesel engine using CFD simulations and a graphical processing unit (GPU)-based chemical kinetics solver (Zero-RK, developed at Lawrence Livermore National Laboratory, Livermore, CA) on high performance computing resources to enable the use of detailed kinetic mechanisms. Diesel engine combustion computations have been conducted over 600 operating points spanning in-vehicle speed-load map, using massively parallel ensemble simulation sets on the Titan supercomputer located at the Oak Ridge Leadership Computing Facility. The results with different mesh resolutions have been analyzed to compare differences in combustion and emissions (NOx, carbon monoxide CO, unburned hydrocarbons (UHC), and smoke) with actual engine measurements. The results show improved agreement in combustion and NOx predictions with a large n-heptane mechanism consisting of 144 species and 900 reactions with refined mesh resolution; however, agreement in CO, UHC, and smoke remains a challenge.

42 ENGINEERING↗

Scaling neural simulations in STACS

Abstract As modern neuroscience tools acquire more details about the brain, the need to move towards biological-scale neural simulations continues to grow. However, effective simulations at scale remain a challenge. Beyond just the tooling required to enable parallel execution, there is also the unique structure of the synaptic interconnectivity, which is globally sparse but has relatively high connection density and non-local interactions per neuron. There are also various practicalities to consider in high performance computing applications, such as the need for serializing neural networks to support potentially long-running simulations that require checkpoint-restart. Although acceleration on neuromorphic hardware is also a possibility, development in this space can be difficult as hardware support tends to vary between platforms and software support for larger scale models also tends to be limited. In this paper, we focus our attention on Simulation Tool for Asynchronous Cortical Streams (STACS), a spiking neural network simulator that leverages the Charm++ parallel programming framework, with the goal of supporting biological-scale simulations as well as interoperability between platforms. Central to these goals is the implementation of scalable data structures suitable for efficiently distributing a network across parallel partitions. Here, we discuss a straightforward extension of a parallel data format with a history of use in graph partitioners, which also serves as a portable intermediate representation for different neuromorphic backends. We perform scaling studies on the Summit supercomputer, examining the capabilities of STACS in terms of network build and storage, partitioning, and execution. We highlight how a suitably partitioned, spatially dependent synaptic structure introduces a communication workload well-suited to the multicast communication supported by Charm++. We evaluate the strong and weak scaling behavior for networks on the order of millions of neurons and billions of synapses, and show that STACS achieves competitive levels of parallel efficiency.

59 BASIC BIOLOGICAL SCIENCES↗

Steady-State Calibration of a Diesel Engine in CFD Using a GPU-Based Chemistry Solver

The prospect of analysis-driven pre-calibration of a modern diesel engine is extremely valuable in order to significantly reduce hardware investments and accelerate engine designs compliant with stricter EPA fuel economy regulations. Advanced modeling tools, such as CFD, are often used with the goal of streamlining significant portions of the calibration process. The success of the methodology largely relies on the accuracy of analytical predictions, especially engine-out emissions. However, the effectiveness of CFD simulation tools for in-cylinder engine combustion is often compromised by the complexity, accuracy, and computational overhead of detailed chemical kinetics necessary for combustion calculations. The standard approach has been to use skeletal kinetic mechanisms (~50 species) which consume acceptable computational time but with degraded accuracy.In this work, a comprehensive demonstration and validation of the analytical pre-calibration process is presented for a passenger car diesel engine using CFD simulations with CONVERGE™ and a GPU-based chemical kinetics solver (Zero-RK, developed at Lawrence Livermore National Laboratory) on high performance computing resources to enable the use of detailed kinetic mechanisms. Diesel engine combustion computations have been conducted over 600 operating points spanning in-vehicle speed-load map, using massively parallel ensemble simulation sets on the Titan supercomputer located at the Oak Ridge Leadership Computing Facility. The results with different mesh resolutions have been analyzed to compare differences in combustion and emissions (NO x , Carbon Monoxide CO, Unburned Hydrocarbons UHC, and Smoke) with actual engine measurements. The results show improved agreement in combustion and NO x predictions with a large n-heptane mechanism consisting of 144 species and 900 reactions with refined mesh resolution; however; agreement in CO, UHC and Smoke remain a challenge.

Gao, Jian↗

Cabana: A Performance Portable Library for Particle-Based Simulations

Particle-based simulations are ubiquitous throughout many fields of computational science and engineering, spanning the atomistic level with molecular dynamics (MD), to mesoscale particle-in-cell (PIC) simulations for solid mechanics, device-scale modeling with PIC methods for plasma physics, and massive N-body cosmology simulations of galaxy structures, with many other methods in between (Hockney & Eastwood, 1989). While these methods use particles to represent significantly different entities with completely different physical models, many low-level details are shared including performant algorithms for short- and/or long-range particle interactions, multi-node particle communication patterns, and other data management tasks such as particle sorting and neighbor list construction. Cabana is a performance portable library for particle-based simulations, developed as part of the Co-Design Center for Particle Applications (CoPA) within the Exascale Computing Project (ECP) (Alexander et al., 2020). The CoPA project and its full development scope, including ECP partner applications, algorithm development, and similar software libraries for quantum MD, is described in (Mniszewski et al., 2021). Cabana uses the Kokkos library for on-node parallelism (Edwards et al., 2014; Trott et al., 2022), enabling simulation on multi-core CPU and GPU architectures, and MPI for GPU-aware, multi-node communication. Cabana provides particle simulation capabilities on almost all current Kokkos backends, including serial execution, OpenMP (including OpenMP-Target for GPUs), CUDA (NVIDIA GPUs), HIP (AMD GPUs), and SYCL (Intel GPUs), providing a clear path for the coming generation of accelerator-based exascale hardware. Cabana builds on Kokkos by providing new particle data structures and particle algorithms resulting in a similar execution policy-based, node-level programming model that is intended to be used in addition to the core Kokkos library within an application. Cabana is designed as an application and physics agnostic, but particle-specific toolkit which can either be used to generate a new application, or to be used as needed in existing applications at various levels of invasiveness including through interfaces that wrap user memory in existing data structures.

97 MATHEMATICS AND COMPUTING↗

Predictive Battery Management System for Commercial Hybrid Vehicles: Cooperative Research and Development Final Report, CRADA Number CRD-13-520

NREL and Eaton Corporation will perform work together on DOE Advanced Research Projects Agency-Energy (ARPA-E) 'Advanced Management and Protection of Energy Storage Devices' program (DE-FOA0000675). NREL will experimentally characterize aging behavior of Eaton cells and packs. Eaton and NREL will implement NREL's prognostic life model in Eaton HEV supervisory controllers and demonstrate the algorithms in accelerated life hardware-in-the loop testing conducted at NREL.

25 ENERGY STORAGE↗

Advanced Module Architecture for Reduced Costs, High Durability and Significantly Improved Manufacturability (Final Report)

This project demonstrated a new module architecture (referred to herein as Glass/Glass) using silicone edge seals and an interlayer polymer that reduces manufacturing costs, significantly streamlines manufacturing processes, and reduces cap-ex costs while improving module reliability. A prototype manufacturing process to fabricate this new module architecture was demonstrated with a significantly improved process cycle time for the lamination step from the current industry standard of 13.5 minutes to approximately 30 seconds for each of the individual edge seal and internal polymer application steps. Samples are fabricated, strenuously stressed, and characterized to quantify the cost benefits, hardware capacities and accelerated stress performance of the new architecture. These results are being compared to a standard, laminated baseline architecture using a Glass/EVA/Glass or Glass/Thermoplastic/Glass design that traditionally has high moisture vapor transmission rates (MVTR).

14 SOLAR ENERGY↗

Porting DMRG++ Scientific Application to OpenPOWER

With the rapidly changing microprocessor designs and architectural diversity (multi-cores, many-cores, accelerators) for the next generation HPC systems, scientific applications must adapt to the hardware, to exploit the different types of parallelism and resources available in the architecture. To get the benefit of all the in-node hardware threads, it is important to use a single programming model to map and coordinate the available work to the different heterogeneous execution units in the node (e.g., multi-core hardware threads (latency optimized), accelerators (bandwidth optimized), etc.).Our goal is to show that we can manage the node complexity of these systems by using OpenMP for in-node parallelization by exploiting different “programming styles” supported by OpenMP 4.5 to program CPU cores and accelerators. Finding out the suitable programming-style (e.g., SPMD style, multi-level tasks, accelerator programming, nested parallelism, or a combination of these) using the latest features of OpenMP to maximize performance and achieve performance portability across heterogeneous and homogeneous systems is still an open research problem.We developed a mini-application, Kronecker Product (KP), from the original DMRG++ application (sparse matrix algebra) computational motif to experiment with different OpenMP programming styles on an OpenPOWER architecture and present their results in this paper.

Chatterjee, Ronnie↗

Coupled Cluster Theory on Graphics Processing Units I. The Coupled Cluster Doubles Method

The coupled cluster (CC) ansatz is generally recognized as providing one of the best wave function-based descriptions of electronic correlation in small- and medium-sized molecules. The fact that the CC equations with double excitations (CCD) may be expressed as a handful of dense matrix-matrix multiplications makes it an ideal method to be ported to graphics processing units (GPUs). Here we present our implementation of the spin-free CCD equations in which the entire iterative procedure is evaluated on the GPU. The GPU-accelerated algorithm readily achieves a factor of 4-5 speedup relative to the multithreaded CPU algorithm on same-generation hardware. The GPU-accelerated algorithm is approximately 8-12 times faster than Molpro, 17-22 times faster than NWChem, and 21-29 times faster than GAMESS for each CC iteration. Single-precision GPU-accelerated computations are also performed, leading to an additional doubling of performance. Single-precision errors in the energy are typically on the order of 10 -6 hartrees and can be improved by about an order of magnitude by performing one additional iteration in double precision.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

TOPOS: A new topometric patient positioning and tracking system for radiation therapy based on structured white light

Purpose: A patient positioning system for radiation therapy based on structured white light and using off-the-shelf hardware components for flexibility and cost-effectiveness has been developed in house. Increased accuracy, patient comfort, abandonment of any skin marks, accelerated workflow, objective reading/recording, better usability and robust sensor design, compared to other positioning approaches, were the main goals of this work. Another aim was the application of a 6 degrees of freedom tracking system working without dose deposition. Methods: Two optical sensors are the main parts of the TOPOS Registered-Sign system (Topometrical Positioning, cyberTECHNOLOGIES, Germany). The components: cameras, projectors, and computers are commercial off-the-shelf products, allowing for low production costs. The black/white cameras of the prototype are capable of taking up to 240 frames per second (resolution: 640 Multiplication-Sign 488 pixels). The projector has a resolution of 1024 Multiplication-Sign 768 and a refresh rate of 120 Hz. The patient's body surface is measured continuously and registered to a reference surface, providing a transformation to superimpose the patient's surface to the reference (planning CT) surface as best as possible. The execution of the calculated transformation provides the correct patient position before the treatment starts. Due to the high-speed acquisition of the surfaces, a surveillance of the patient's (respiration) motion during treatment is also accomplished. The accuracy of the system was determined using a male mannequin. Two treatment sites were evaluated: one simulating a head and neck treatment and the other simulating a thoracic wall treatment. The mannequin was moved to predefined positions, and shift vectors given by the surface registration were evaluated. Additionally manual positioning using a color-coding system was evaluated. Results: Two prototypes have been developed, each allowing a continuous high density scan of a 500 Multiplication-Sign 500 Multiplication-Sign 400 mm{sup 3} (L Multiplication-Sign W Multiplication-Sign D) large volume with a refresh rate of 10 Hz (extendible to 20 Hz for a single sensor system). Surface and position correction display, as well as respiratory motion, is shown in real-time (delay < 200 ms) using present graphical hardware acceleration. For an intuitive view of the patient's misalignment, a fast surface registration algorithm has been developed and tested and a real-time color-coding technique is proposed and verified that allows the user to easily verify the position of the patient. Using first the surface registration and then the color coding the best results were obtained: for the head and neck case, the mean difference between the actual zero position and the final match was 0.1 {+-} 0.4, -0.2 {+-} 0.7, and -0.1 {+-} 0.3 mm in vertical, longitudinal, and lateral direction. For the thoracic case, the mean differences were 0.3 {+-} 0.5, -0.6 {+-} 1.9, 0.0 {+-} 0.4 mm. Conclusions: The presented system copes with the increasing demand for more accurate patient positioning due to more precise irradiation technologies and minimizes the preparation times for correct patient alignment, therefore optimizing the treatment workflow. Moreover, TOPOS is a versatile and cost effective image guided radiation therapy device. It allows an objective rating of the patient's position before and during the irradiation and could also be used for respiratory gating or tracking.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Heavy ion beam physics at Facility for Rare Isotope Beams

The Facility for Rare Isotope Beams (FRIB) will be the world's premier rare-isotope beam facility. Experiments with the majority (~80%) of the isotope predicted to exist will become available. The FRIB facility is based on a superconducting (SC) heavy ion linac with output energy above 200 MeV/u for any ions at beam power of 400 kW. FRIB includes a target facility for in-flight production of rare isotopes. A three-stage fragment separator will be used to prepare fast rare isotope beams with high-purity for nuclear physics experiments. The installation work of the accelerator and experimental systems is approaching completion and multi-stage beam commissioning activities started in summer 2017 with expected project completion in early 2022. In conclusion, the commencement of operation for users' experiments is planned immediately following the project completion.

43 PARTICLE ACCELERATORS↗

AI-Powered Knowledge Graphs for Neuromorphic and Energy-Efficient Computing

The surge in scientific literature obscures breakthroughs and hinders the discovery of new research paths. We propose an artificial intelligence (AI) powered framework using large language models (LLMs) and knowledge graphs (KGs) to automate parts of scientific discovery, focusing on energy-efficient AI circuits. Our hybrid approach combines LLMs, structured data, and ontology-based reasoning to construct a comprehensive knowledge graph that integrates insights across computational neuroscience, spiking neuron models, learning rules, architectural motifs, and neuromorphic device technologies. This multi-domain representation enables the generation of hypotheses that connect biological function with implementable, energy-efficient hardware architectures. Using KG embeddings and graph neural networks, the framework generates hypotheses for novel circuits, validates them through optimization on exascale HPC systems, and with tools like SuperNeuro and Fugu, the most promising designs will be prototyped in hardware. This open-source system aims to accelerate discoveries and bridging neuroscience with hardware innovation, drive collaboration, and unlock new opportunities in low-power AI computing.

Gautam, Ashish [ORNL]↗

Shifting Between Compute and Memory Bounds: A Compression-Enabled Roofline Model

In the evolving landscape of high-performance computing, especially to fight the end of Moore’s Law and Dennard’s Scaling, the ability to shift between compute-bound and memory-bound states is critical for enhancing adaptability and flexibility to diverse system and domain-specific architectures. Such capability is vital for optimizing performance across distinguished hardware configurations, such as accelerators, memory hierarchies, and cache systems. Despite that ad hoc optimization techniques, such as compressed/approximate computation, have been enabled for compute-/data-intensive computing for improved performance in distinct hardware settings, there lacks an understanding of 1) the rational behind performance improvement; 2) capability of different optimizations; 3) what optimization to respond to specific computational and memory demands. This work proposes a compression-enabled roofline model to facilitate this adaptability with data compression techniques to balance and transform between computational and memory demands. This model enables applications to adjust in response to the specific strengths and limitations of the underlying hardware and system to optimize resource utilization. The effectiveness of this approach is demonstrated with matrix multiplication kernels on different input sizes, with turning on/off various compression techniques, including 1) low-precision floating point; 2) sparse matrix formulation; and 3) compressed arrays with ZFP. By reducing memory transfer volumes and cache misses and increasing data locality and computational intensity through compression, the specific roofline model can transform between compute and memory bounds to align more efficiently with system capabilities. This advancement not only improves overall performance but also maximizes adaptability in diverse computing environments.

Naraparaju, Ramasoumya [University of Washington]↗

Tournament-Based Pretraining to Accelerate Federated Learning

Advances in hardware, proliferation of compute at the edge, and data creation at unprecedented scales have made federated learning (FL) necessary for the next leap forward in pervasive machine learning. For privacy and network reasons, large volumes of data remain stranded on endpoints located in geographically austere (or at least austere network-wise) locations. However, challenges exist to the effective use of these data. To solve the system and functional level challenges, we present an three novel variants of a serverless federated learning framework. We also present tournament-based pretraining, which we demonstrate significantly improves model performance in some experiments. Overall, these extensions to FL and our novel training method enable greater focus on science rather than ML development.

Baughman, Matt↗

Enabling real-time adaptation of machine learning models at x-ray Free Electron Laser facilities with high-speed training optimized computational hardware

The emergence of novel computational hardware is enabling a new paradigm for rapid machine learning model training. For the Department of Energy’s major research facilities, this developing technology will enable a highly adaptive approach to experimental sciences. In this manuscript we present the per-epoch and end-to-end training times for an example of a streaming diagnostic that is planned for the upcoming high-repetition rate x-ray Free Electron Laser, the Linac Coherent Light Source-II. We explore the parameter space of batch size and data parallel training across multiple Graphics Processing Units and Reconfigurable Dataflow Units. We show the landscape of training times with a goal of full model retraining in under 15 min. Although a full from scratch retraining of a model may not be required in all cases, we nevertheless present an example of the application of emerging computational hardware for adapting machine learning models to changing environments in real-time, during streaming data acquisition, at the rates expected for the data fire hoses of accelerator-based user facilities.

97 MATHEMATICS AND COMPUTING↗