SEARCH · Engineering Papers

Results for “inference 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.

SourceResearch connectionFrom yearThrough year

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

At least 289 records · Page 16

All sources NASA NTRS DOE OSTI

DOE OSTI2024

Optimizing High-Throughput Inference on Graph Neural Networks at Shared Computing Facilities with the NVIDIA Triton Inference Server

Abstract With machine learning applications now spanning a variety of computational tasks, multi-user shared computing facilities are devoting a rapidly increasing proportion of their resources to such algorithms. Graph neural networks (GNNs), for example, have provided astounding improvements in extracting complex signatures from data and are now widely used in a variety of applications, such as particle jet classification in high energy physics (HEP). However, GNNs also come with an enormous computational penalty that requires the use of GPUs to maintain reasonable throughput. At shared computing facilities, such as those used by physicists at Fermi National Accelerator Laboratory (Fermilab), methodical resource allocation and high throughput at the many-user scale are key to ensuring that resources are being used as efficiently as possible. These facilities, however, primarily provide CPU-only nodes, which proves detrimental to time-to-insight and computational throughput for workflows that include machine learning inference. In this work, we describe how a shared computing facility can use the NVIDIA Triton Inference Server to optimize its resource allocation and computing structure, recovering high throughput while scaling out to multiple users by massively parallelizing their machine learning inference. To demonstrate the effectiveness of this system in a realistic multi-user environment, we use the Fermilab Elastic Analysis Facility augmented with the Triton Inference Server to provide scalable and high-throughput access to a HEP-specific GNN and report on the outcome.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

NASA NTRS1998

Central Plasma Sheet Ion Properties as Inferred from Ionospheric Observations

A method of inferring central plasma sheet (CPS) temperature, density, and pressure from ionospheric observations is developed. The advantage of this method over in situ measurements is that the CPS can be studied in its entirely, rather than only in fragments. As a result, for the first time, comprehensive two-dimensional equatorial maps of CPS pressure, density, and temperature within the isotropic plasma sheet are produced. These particle properties are calculated from data taken by the Special Sensor for Precipitating Particles, version 4 (SSJ4) particle instruments onboard DMSP F8, F9, F10, and F11 satellites during the entire year of 1992. Ion spectra occurring in conjunction with electron acceleration events are specifically excluded. Because of the variability of magnetotail stretching, the mapping to the plasma sheet is done using a modified Tsyganenko [1989] magnetic field model (T89) adjusted to agree with the actual magnetotail stretch at observation time. The latter is inferred with a high degree of accuracy (correlation coefficient -0.9) from the latitude of the DMSP b2i boundary (equivalent to the ion isotropy boundary). The results show that temperature, pressure, and density all exhibit dawn-dusk asymmetries unresolved with previous measurements. The ion temperature peaks near the midnight meridian. This peak, which has been associated with bursty bulk flow events, widens in the Y direction with increased activity. The temperature is higher at dusk than at dawn, and this asymmetry increases with decreasing distance from the Earth. In contrast, the density is higher at dawn than at dusk, and there appears to be a density enhancement in the low-latitude boundary layer regions which increases with decreasing magnetic activity. In the near-Earth regions, the pressure is higher at dusk than at dawn, but this asymmetry weakens with increasing distance from the Earth and may even reverse so that at distances X less than approx. 10 to -12 R(sub E), depending on magnetic activity, the dawn sector has slightly higher pressure. The temperature and density asymmetries in the near-Earth region are consistent with the ion westward gradient/curvature drift as the ions ExB convect earthward. When the solar wind dynamic pressure increases, CPS density and pressure appear to increase, but the temperature remains relatively constant. Comparison with previously published work indicates good agreement between the inferred pressure, temperature, and density and those obtained from in situ data. This new method should provide a continuous mechanism to monitor the pressure, temperature, and density in the magnetotail with unprecedented comprehensiveness.

Wing, Simon↗

NASA NTRS1997

Cardiovascular responses of snakes to hypergravity

Snakes have provided useful vertebrate models for understanding circulatory adaptation to gravity, attributable to their elongate body shape and evolutionary diversificaton in terms of ecology and behavior. Recently we have studied cardiovascular responses of snakes to hypergravic acceleration forces produced acutely in the head-to-tail direction (+Gz) on a short-arm centrifuge. Snakes were held in a nearly straight position within a horizontal plastic tube and subjected to a linear force gradient during acceleration. Carotid blood flow provided an integrated measure of cardiovascular performance. Thus, cardiovascular tolerance of snakes to stepwise increments of Gz was measured as the caudal Gz force at which carotid blood flow ceased. Tolerance to increasing Gz varies according to adaptive evolutionary history inferred from the ecology and behavior of species. With respect to data for six species we investigated, multiple regression analysis demonstrates that Gz tolerance correlates with gravitational habitat, independently of body length. Relative to aquatic and non-climbing species, carotid blood flow is better maintained in arboreal or scansorial species, which tolerate hypergravic forces of +2 to +3.5 Gz. Additionally, semi-arboreal rat snakes (Elaphe obsoleta) exhibit plasticity of responses to long-term, intermittent +1.5 Gz stress. Compared to non-acclimated controls, acclimated snakes show greater increases of heart rate during head-up tilt or acceleration, greater sensitivity of arterial pressure to circulating catecholamines, higher blood levels of prostaglandin ratios favorable to maintenance of arterial blood pressure, and medial hypertrophy in major arteries and veins. As in other vertebrates, Gz tolerance of snakes is enhanced by acclimation, high arterial pressure, comparatively large blood volume, and body movements. Vascular studies of snakes suggest the importance to acclimation of local responses involving vascular tissue, in addition to centrally mediated responses to fluid shifts.

Review, Tutorial↗

DOE OSTI2025

Fast Adaptive Neural Control of Resonant Extraction at Fermilab

We present the development of a machine learning (ML) based regulation system for third-order resonant beam extraction in the Mu2e experiment at Fermilab. Classical and ML-based controllers have been optimized using semi-analytic simulations and evaluated in terms of regulation performance and training efficiency. We compare several controller architectures and discuss the integration of neural control into an adaptive framework. We also present progress on surrogate models that predict the controller response given a spill intensity and controller action history. To enable real-time deployment, we report progress on implementing low-latency, edge-based inference suitable for hardware-constrained environments. Our results demonstrate the feasibility and advantages of ML-based control in managing complex, time-varying physical systems, with broader implications for accelerator operations and other domains requiring fast, adaptive regulation.

Berlioz, Jose Rene [Fermilab]↗

NASA NTRS1973

Coronal propagation of energetic charged particles

A technique for estimating the equatorial connection longitude of interplanetary field lines to the corona is applied to quasi-stationary low energy solar charged particle (LESP) events to reconstruct their coronal injection profile. Comparison with synoptic charts constructed from alpha chromospheric absorption features for solar rotation 1524 reveals that the inferred chromospheric equatorial magnetic neutral line structure strikingly orders the LESP coronal injection profiles of greater than or equal 0.3 MeV protons for this period (August 1967), allowing the identification of acceleration processes lasting several days and well defined coronal transport over greater than or approximately 100 deg. It is also proposed that LESP observations offer a unique tool for studying the magnetic structure of the transition region from the low corona to interplanetary space.

Roelof, E. C.↗

NASA NTRS1978

Remote in-situ elemental analysis systems for underwater application

The systems approach, theoretical measurement calculations, and preliminary measurements to be used in monitoring and mapping pollutants (such as traces of heavy metals) in the Chesapeake Bay are discussed. A neutron gamma-ray method is under development for demonstrating the system. The excitation source to be used is a machine accelerator using a deuterium/tritium reaction to produce 14-MeV neutrons. The neutrons excite characteristic gamma ray emission from the neutron irradiated surface. The discrete line emission produced can be used to infer both qualitative and quantitative elemental composition. A data preprocessor will accumulate, digitize, store, format, and prepare the data for transmission which can be accomplished by telephone, microwave, and possibly satellite link to central processors.

Trombka, J. I.↗

DOE OSTI2024

Analyzing inference workloads for spatiotemporal modeling

Ensuring power grid resiliency, forecasting climate conditions, and optimization of transportation infrastructure are some of the many application areas where data is collected in both space and time. Spatiotemporal modeling is about modeling those patterns for forecasting future trends and carrying out critical decision-making by leveraging machine learning/deep learning. Once trained offline, field deployment of trained models for near real-time inference could be challenging because performance can vary significantly depending on the environment, available compute resources and tolerance to ambiguity in results. Users deploying spatiotemporal models for solving complex problems can benefit from analytical studies considering a plethora of system adaptations to understand the associated performance-quality trade-offs. To facilitate the co-design of next-generation hardware architectures for field deployment of trained models, it is critical to characterize the workloads of these deep learning (DL) applications during inference and assess their computational patterns at different levels of the execution stack. In this paper, we develop several variants of deep learning applications that use spatiotemporal data from dynamical systems. We study the associated computational patterns for inference workloads at different levels, considering relevant models (Long short-term Memory, Convolutional Neural Network and Spatio-Temporal Graph Convolution Network), DL frameworks (Tensorflow and PyTorch), precision (FP16, FP32, AMP, INT16 and INT8), inference runtime (ONNX and AI Template), post-training quantization (TensorRT) and platforms (Nvidia DGX A100 and Sambanova SN10 RDU). Overall, our findings indicate that although there is potential in mixed-precision models and post-training quantization for spatiotemporal modeling, extracting efficiency from contemporary GPU systems might be challenging. Instead, co-designing custom accelerators by leveraging optimized High Level Synthesis frameworks (such as SODA High-Level Synthesizer for customized FPGA/ASIC targets) can make workload-specific adjustments to enhance the efficiency.

97 MATHEMATICS AND COMPUTING↗

NASA NTRS1978

Stochastic acceleration of solar flare protons

The acceleration of solar flare protons is considered by cyclotron damping of intense Alfven wave turbulence in a magnetic trap. The energy diffusion coefficient is computed for a near-isotropic distribution of super-Alfvenic protons and a steady-state solution for the particle spectrum is found for both transit-time and diffusive losses out of the ends of the trap. The acceleration time to a characteristic energy approximately 20 Mev/nucl can be as short as 10 sec. On the basis of phenomenological arguments an omega/2 frequency dependence for the Alfven wave spectrum is inferred. The correlation time of the turbulence lies in the range .0005 less than tau/corr less than .05s.

Barbosa, D. D.↗

NASA NTRS2004

External and Internal Reconnection in Two Filament-Carrying Magnetic-Cavity Solar Eruptions

We observe two near-limb solar filament eruptions, one of 2000 February 26 and the other of 2002 January 4. For both we use 195 Angstroms, Fe XII images from the Extreme-Ultraviolet Imaging Telescope (EIT) and magnetograms from the Michelson Doppler Imager (MDI), both of which are on the Solar arid Heliospheric Observatory (SOHO). For the earlier event we also use soft X-ray telescope (SXT), hard X-ray telescope (HXT), and Bragg Crystal Spectrometer (BCS) data from the Yohkoh satellite, and hard X-ray data from the BATSE experiment on the Compton Gamma Ray Observation, (CGRO). Both events occur in quadrupolar magnetic regions, and both have coronal features that we infer belong to the same magnetic cavity structures as the filaments. In both cases, the cavity and filament first rise slowly at approximately 10 kilometers per second prior to eruption and then accelerate to approximately 100 kilometers per second during the eruption, although the slow-rise movement for the higher altitude cavity elements is clearer in the later event. We estimate that both filaments and both cavities contain masses of approximately 10(exp 14)-10(exp 15) and approximately 10(exp 15)-10(exp 16)g, respectively. We consider whether two specific magnetic reconnection-based models for eruption onset, the tether cutting and the breakout models, are consistent with our observations. In the earlier event, soft X-rays from SXT show an intensity increase during the 12 minute interval over which fast eruption begins, which is consistent with tether-cutting-model predictions. Substantial hard X-rays, however, do not occur until after fast eruption is underway, and so this is a constraint the tether-cutting model must satisfy. During the same 12 minute interval over which fast eruption begins, there are brightenings and topological changes in the corona indicative of high-altitude reconnection early in the eruption, and this is consistent with breakout predictions. In both eruptions, the state of the overlying loops at the time of onset of the fast-rise phase of the corresponding filament can be compared with expectations from the breakout model, thereby setting constraints that the breakout model must meet. Our findings are consistent with both runaway tether-cutting-type reconnection and fast breakout-type reconnection, occurring early in the fast phase of the February eruption and with both types of reconnection being important in unleashing the explosion, but we are not able to say which, if either, type of reconnection actually triggered the fast phase. In any case, we have found specific constraints that either model, or any other model, must satisfy if correct.

Sterling, Alphonse C.↗

NASA NTRS2012

Inner Plasma Structure of the Low-Latitude Reconnection Layer

We report a clear transition through a reconnection layer at the low-latitude magnetopause which shows a complete traversal across all reconnected field lines during northwestward interplanetary magnetic field (IMF) conditions. The associated plasma populations confirm details of the electron and ion mixing and the time history and acceleration through the current layer. This case has low magnetic shear with a strong guide field and the reconnection layer contains a single density depletion layer on the magnetosheath side which we suggest results from nearly field-aligned magnetosheath flows. Within the reconnection boundary layer, there are two plasma boundaries, close to the inferred separatrices on the magnetosphere and magnetosheath sides (Ssp and Ssh) and two boundaries associated with the Alfvén waves (or Rotational Discontinuities, RDsp and RDsh). The data are consistent with these being launched from the reconnection site and the plasma distributions are well ordered and suggestive of the time elapsed since reconnection of the field lines observed. In each sub-layer between the boundaries the plasma distribution is different and is centered around the current sheet, responsible for magnetosheath acceleration. We show evidence for a velocity dispersion effect in the electron anisotropy that is consistent with the time elapsed since reconnection. In addition, new evidence is presented for the occurrence of partial reflection of magnetosheath electrons at the magnetopause current layer.

magnetosheath↗

NASA NTRS1979

Application of the trap-plus-precipitation hard X-ray burst model to the flare of August 4, 1972

The theory of Melrose and Brown (1976) concerning the X-ray spectrum resulting from combined precipitation and in-trap acceleration of nonthermal electrons is applied to a large solar flare on the basis of high-time-resolution hard X-ray data. The time behavior of the electron source function is inferred as a function of energy and mean ambient trap density. The results are interpreted in the context of a purely thick-target model, a static trap with continuous injection of fresh electrons, and a dynamic trap with continuous redistribution of electron energies. It is concluded that: (1) a minimum trap density exists below which the static-trap description cannot be valid; (2) a mean ambient trap density less than the cited minimum is possible if a dynamic trap is invoked; and (3) the electron source function at a given instant in such low-density dynamic traps may bear little or no resemblance to a power-law spectrum despite its production, via nonthermal bremsstrahlung, of a power-law X-ray spectrum.

Emslie, A. G.↗

NASA NTRS1995

VLF waves in the foreshock

Plasma waves observed in the VLF range upstream of planetary bow shocks not only modify the particle distributions, but also provide important information about the acceleration processes that occur at the bow shock. Electron plasma oscillations observed near the tangent field line in the electron foreshock are generated by electrons reflected at the bow shock through a process that has been referred to as Fast Fermi acceleration. Fast Fermi acceleration is the same as shock-drift acceleration, which is one of the mechanisms by which ions are energized at the shock. We have generated maps of the VLF emissions upstream of the Venus bow shock, using these maps to infer properties of the shock energization processes. We find that the plasma oscillations extend along the field line up to a distance that appears to be controlled by the shock scale size, implying that shock curvature restricsts the flux and energy of reflected electrons. We also find that the ion acoustic waves are observed in the ion foreshock, but at Venus these emissions are not detected near the ULF forshock boundary. Through analogy with terrestrial ion observations, this implies that the ion acoustic waves are not generated by ion beams, but are instead generated by diffuse ion distributions found deep within the ion foreshock. However, since the shock is much smaller at Venus, and there is no magnetosphere, we might expect ion distributions within the ion foreshock to be different than at the Earth. Mapping studies of the terrestrial foreshock similar to those carried out at Venus appear to be necessary to determine if the inferences drawn from Venus data are applicable to other foreshocks.

Strangeway, R. J.↗

NASA NTRS2004

External and Internal Reconnection in Two Filament-Carrying Magnetic-Cavity Solar Eruptions

We observe two near-limb solar filament eruptions, one of 2000 February 26 and the other of 2002 January 4. For both we use 195 Angstroms Fe XII images from the Extreme-Ultraviolet (EUV) Imaging Telescope (EIT) and magnetograms from the Michelson Doppler Imager (MDI), both on the Solar and Heliospheric Observatory (SOHO) satellite. For the earlier event we also use soft X-ray telescope (SXT), hard X-ray telescope (HXT), and Bragg Crystal Spectrometer (BCS) data from the Yohkoh satellite, and hard X-ray data from the BATSE experiment on the Compton Gamma Ray Observatory (CGRO). Both events occur in quadrupolar magnetic regions, and both have coronal features that we infer belong to the same magnetic-cavity structures as the filaments. In both cases the cavity and filament first rise slowly at approximately 10 kilometers per second prior to eruption, and then accelerate to approximately 100 kilometers per second during the eruption, although the slow-rise movement for the higher-altitude cavity elements is clearer in the later event. We estimate both filaments and both cavities to contain masses of approximately 10(exp 14-15) g and approximately 10(exp 15-16) g respectively. We consider whether two specific magnetic-reconnection-based models for eruption onset, the tether cutting and the breakout models, are consistent with our observations. In the earlier event soft X-rays from SXT show an intensity increase during the 12-minute interval over whch fast eruption begins, which is consistent with tether-cutting-model predictions. Substantial hard X-rays, however, do not occur until after fast eruption is underway, and so this is a constraint the tether-cutting model must satisfy. During the same 12-minute interval over which fast eruption begins, there are brightenings and topological changes in the corona indicative of high-altitude reconnection early in the eruption, and this is consistent with breakout predictions. In both eruptions, however, the onset of the filament fast-rise phase occurs while overlying cavity-related coronal loops are still evolving from a "closed" state to an "open" state, and our observations of the time evolution of these loops compared to the eruption state in both events are constraints the breakout model must meet. Therefore our findings are consistent with runaway-tether-cutting-type reconnection and fast breakout-type reconnection both occurring early in the fast phase of the February eruption and with both types of reconnection being important in unleashing the explosion, but we are not able to say which, if either, type of reconnection actually triggered the fast phase. We have also found specific constraints that either model, or any other model, must satisfy if correct.

Sterling, Alphonse C.↗

NASA NTRS1986

ISEE observations of the plasma sheet boundary, plasma sheet, and neutral sheet. I - Electric field, magnetic field, plasma, and ion composition

The first simultaneous study of dc and ac electric and magnetic fields, E x B velocity, plasma flows, ratio of plasma to magnetic field pressure, total energy density, energetic particles, and ion composition from the ISEE satellites and ground and interplanetary magnetic fields has been made to determine (1) the relationship of the previously observed electric fields at the plasma sheet boundary and at the neutral sheet to plasma parameters, and (2) whether the phenomena occurring during quiet and active times were consistent with the formation of a near-earth neutral line during substorms or with the boundary layer model. Five observations made during the study of two substorms were seen to be in agreement with the neutral-line model. The observations are consistent with the satellite being located at varying distances from the neutral line and diffusion region where reconnection and plasma acceleration were occurring. Although the z component (into or out of the ecliptic plane) of E x B convection was generally toward the neutral sheet, there were examples when it was consistent with the inferred motion of the plasma sheet past the satellite. A synthesis of previous reports on large electric fields at the plasma sheet boundary and variable fields at the neutral sheet including the associated plasma flows is also described.

Cattell, C. A.↗

NASA NTRS1987

Beamed emission from gamma-ray burst sources

Gamma-ray bursts are intense fluxes of radiation in the 100 keV to several MeV energy range which typically persist for between a fraction of a second to several seconds. The observed spectral shape of these bursts suggest that the radiation is emitted as highly collimated beams emanating from neutron stars. This inference is based on the lack of significant gamma-gamma absorption (which are produced when gamma rays interact with stellar surfaces). The gamma-ray beams may be a consequence of a particle acceleration in double layers in neutron star magnetospheres.

Epstein, R.↗

NASA NTRS1990

Simulation studies of electron acceleration by ion ring distributions in solar flares

A 2.5-dimensional fully relativistic EM, particle-in-cell code (PIC) is used to investigate a potential electron acceleration mechanism in solar flares. The free energy is provided by ions which have a ring velocity distribution about the magnetic-field direction. Ion rings may be produced by perpendicular shocks, which could in turn be generated by the super-Alfvenic motion of magnetic flux tubes emerging from the photosphere or by coronal mass ejections. Such ion distributions are known to be unstable to the generation of lower hybrid waves, which have phase velocities in excess of the electron thermal speed parallel to the field and can, therefore, resonantly accelerate electrons in that direction. The simulations show the transfer of perpendicular ion energy to energetic electrons via lower hybrid wave turbulence. With plausible ion ring velocities, the process can account for the observationally inferred fluxes and energies of non-thermal electrons during the impulsive phase of flares.

Mcclements, K. G.↗

NASA NTRS1995

The He-3/He-4 ratios for solar energetic particle events during the Combined Release and Radiation Effects Satellite Mission

Helium data measured by the University of Chicago instrument, ONR-604, are employed to determine the ratio of He-3 to He-4 for solar energetic particle (SEP) events over an energy range 50-110 MeV/nucleon during the 1990/1991 Combined Release and Radiation Effects Satellite mission. The Sun in this period is extremely active. A total of 29 separate SEP events have been identified; among them 16 major events have been analyzed to obtain He-3/He-4 ratios, with a mass resolution of 0.10 amu. Thirteen events have a He-3/He-4 ratio larger than 0.005, one order of magnitude greater than the solar coronal value. The He-3/He-4 ratio at energies of 50-110 MeV/nucleon is independent of the size of the SEP event, for the moderately large flares analyzed here. The helium energy spectra are represented by power laws. During the 1991 June flare period, different large-particle injections associated with different solar flares, but occurring from the same active region, have a similar average spectral index and a similar He-3/He-4 ratio. The spectral index of He-4 varies from event to event, i.e., from as small as 1.5 to as large as 7.5. A correlation is found between the inferred spectral index from gamma-ray measurements and our measured spectral indices for the 1991 June 11 and June 15 events, suggesting that the high-energy SEPs may come from the same acceleration event as the particles that interact at the Sun and produce the gamma rays. The implications of these results for particle acceleration and propagation at the flare site and in the solar corona are discussed.

Chen, Jiasheng↗

DOE OSTI2022

Atacama Cosmology Telescope: Constraints on prerecombination early dark energy

The early dark energy (EDE) scenario aims to increase the value of the Hubble constant (H0) inferred from cosmic microwave background (CMB) data over that found in the standard cosmological model (Λ CDM ), via the introduction of a new form of energy density in the early Universe. The EDE component briefly accelerates cosmic expansion just prior to recombination, which reduces the physical size of the sound horizon imprinted in the CMB. Previous work has found that nonzero EDE is not preferred by Planck CMB power spectrum data alone, which yield a 95% confidence level (C.L.) upper limit f EDE < 0.087 on the maximal fractional contribution of the EDE field to the cosmic energy budget. In this paper, we fit the EDE model to CMB data from the Atacama Cosmology Telescope (ACT) data release 4. We find that a combination of ACT, large-scale Planck TT (similar to WMAP), Planck CMB lensing, and BAO data prefers the existence of EDE at >99.7 % C .L . : f EDE = 0.091$^{+0.020}_{-0.036}$, with H 0 = 70.9$^{+1.0}_{-2.0}$ km/s/Mpc (both 68% C.L.). From a model-selection standpoint, we find that EDE is favored over Λ CDM by these data at roughly 3 σ significance. In contrast, a joint analysis of the full Planck and ACT data yields no evidence for EDE, as previously found for Planck alone. We show that the preference for EDE in ACT alone is driven by its TE and EE power spectrum data. The tight constraint on EDE from Planck alone is driven by its high-ℓ TT power spectrum data. Understanding whether these differing constraints are physical in nature, due to systematics, or simply a rare statistical fluctuation is of high priority. Here, the best-fit EDE models to ACT and Planck exhibit coherent differences across a wide range of multipoles in TE and EE, indicating that a powerful test of this scenario is anticipated with near-future data from ACT and other ground-based experiments.

79 ASTRONOMY AND ASTROPHYSICS↗