Search NASA⌕ Search

SEARCH · Search NASA

Results for “deep generative models”

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 1,405 records · Page 78

Weak structure functions in $\nu_l-N$ and $\nu_l-A$ scattering with nonperturbative and higher order perturbative QCD effects

We study the effect of various perturbative and nonperturbative QCD corrections on the free nucleon structure functions [FiNWI(x,Q2);i=1–3] and their implications in the determination of nuclear structure functions. The evaluation of the nucleon structure functions has been performed by using the MMHT 2014 parton distribution functions (PDFs) parametrization, and the target mass correction (TMC) and higher twist (HT) effects are incorporated following the works of Schienbein et al. and Dasgupta et al., respectively. These nucleon structure functions are taken as input in the determination of nuclear structure functions. The numerical calculations for the νl/ν¯l-A deep inelastic scattering (DIS) process have been performed by incorporating the nuclear medium effects like Fermi motion, binding energy, nucleon correlations, mesonic contributions, shadowing, and antishadowing in several nuclear targets such as carbon, polystyrene scintillator, iron, and lead, which are being used in MINERνA, and in argon nuclei, which is relevant for the ArgoNeuT and DUNE experiments. The differential scattering cross sections d2σAWIdxdy and (dσAWIdx/dσCHWIdx) have also been studied in the kinematic region of the MINERνA experiment. The theoretical results are compared with the recent experimental data of MINERνA and the earlier data of the NuTeV, CCFR, CDHSW, and CHORUS Collaborations. Moreover, a comparative analysis of the present results for the ratio (dσAWIdx/dσCHWIdx), and the results from the Monte Carlo (MC) generator GENIE and other phenomenological models of Bodek and Yang, and Cloet et al., has been performed in the context of the MINERνA experiment. The predictions have also been made for the ν¯l-A cross section relevant for the MINERνA experiment.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Electricity Pricing aware Deep Reinforcement Learning based Intelligent HVAC Control

Recently, deep reinforcement learning (DRL) based intelligent control of Heating, Ventilation, and Air Conditioning (HVAC) has gained a lot of attention due to DRL's ability to optimally control HVAC for minimizing operational cost while maintaining resident's comfort. The success of such DRL-based techniques largely depends on the articulation of the problem in terms of states, actions, and reward function. Inclusion of the electricity pricing information in the problem formulation can play an important role in saving the cost of HVAC operation. However, less attention has been given in the literature on formulating well-crafted state features based on electricity pricing. In this work, we propose an approach for training the DRL model with a specific focus on feature engineering based on electricity pricing. During training, we generate random but sufficiently realistic electricity price signals so that the pre-trained DRL model is robust and adaptive to the dynamic and variable electricity prices. The validation results are encouraging and show the potential of ≈12%-15% savings in the one day cost of HVAC operation, proving the usefulness of including electricity pricing related features as state features.

Kurte, Kuldeep↗

SCALE Enhancements for Detailed Cask Dose Rate Analysis - Paper 37

The Nuclear Fuels Storage and Transportation Planning Project and the Used Fuel Disposition Campaign, both under the US Department of Energy (DOE) Office of Nuclear Energy, require the ability to predict k{sub eff}, heat generation and temperature distributions, and external dose rates from any combination of commercial spent nuclear fuel assemblies loaded into any type of licensed cask. Work in the Reactor and Nuclear Systems Division at Oak Ridge National Laboratory has begun to develop the UNF-ST[and]DARDS system (Used Nuclear Fuel Storage, Transportation and Disposal Analysis Resource and Data System) that uses many of the nuclear analysis codes from the SCALE package, a comprehensive database of used nuclear fuel inventories from utilities, and a library of model template files to produce complete input files for the various types of UNF computational analyses. UNF-ST[and]DARDS uses the template files and input parameters that are applicable to a safety analysis to assemble and execute a complete input file, collect the output files and relevant information for subsequent calculations, and illustrate calculation results as a function of decay time and other relevant analysis parameters. This paper will focus on the shielding aspects and the ability to predict external dose rates from as-loaded casks. Realistic dose rate values based on actual canister fuel contents may be used in developing appropriate plans, procedures, and methods to keep personnel radiation exposures as low as reasonably achievable during cask repackaging and transportation operations, as well as in performing total system analyses and transportation risk evaluations. The MAVRIC shielding sequence of the SCALE package was created for the US Nuclear Regulatory Commission for analyzing dose rates outside of storage, transfer, or transportation casks loaded with spent nuclear fuel. MAVRIC contains automated variance reduction capabilities based on the CADIS and FW-CADIS methods that are effective for optimizing deep penetration Monte Carlo shielding problems for achieving low relative uncertainties in short computing times. MAVRIC uses the Denovo discrete ordinates code to compute importance maps and biased source terms for the Monaco fixed-source Monte Carlo code, which can use multi-group or continuous-energy cross section data. ORIGEN (Oak Ridge Isotope Generation and Depletion) is the SCALE computer code for calculating time-dependent isotopic concentrations during irradiation and decay for more than 2,200 nuclides that can be produced by nuclear fuel irradiation and activation. A series of SCALE enhancements was implemented to enable automatic decay heat and radiation source term generation and shielding safety analysis of as-loaded casks. The enhancements to SCALE made under this project will be a part of the SCALE 6.2 release expected in fall 2014. (authors)

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

Accidental supersymmetric dark matter and baryogenesis

We show that ''accidental'' supersymmetry is a beyond-the-Standard Model framework that naturally accommodates a thermal relic dark matter candidate and successful electroweak baryogenesis, including the needed strongly first-order character of the electroweak phase transition. We study the phenomenology of this setup from the standpoint of both dark matter and baryogenesis. For energies around the electroweak phase transition temperature, the low-energy effective theory is similar to the MSSM with light super-partners of the third-generation quarks and of the Higgs and gauge bosons. We calculate the dark matter relic abundance and the baryon asymmetry across the accidental supersymmetry parameter space, including resonant and non-resonant CP-violating sources. We find that there are regions of parameter space producing both the observed value of the baryon asymmetry and a dark matter candidate with the correct relic density and conforming to present-day constraints from dark matter searches. This scenario makes sharp predictions for the particle spectrum, predicting a lightest neutralino mass between 200 and 500 GeV, with all charginos and neutralinos within less than a factor 2 of the lightest neutralino mass and the heavy Higgs sector within 20–25% of that mass, making it an interesting target for collider searches. In addition, we demonstrate that successful accidental supersymmetric dark matter and baryogenesis will be conclusively tested with improvements smaller than one order of magnitude to the current performance of electron electric dipole moment searches and of direct dark matter searches, as well as with IceCube plus Deep Core neutrino telescope data.

79 ASTRONOMY AND ASTROPHYSICS↗

Progress in Understanding Short-Range Structure in Nuclei: An Experimental Perspective

High-energy electron scattering is a clean, precise probe for measurements of hadronic and nuclear structure and plays a key role in understanding the role of high-momentum nucleons (and quarks) in nuclei. Jefferson Lab has dramatically expanded our knowledge of the high-momentum nucleons generated by short-range correlations, providing sufficient insight to model much of their impact on nuclear structure in neutron stars and in low- to medium-energy scattering observables, including neutrino oscillation measurements. These short-range correlations also seem related to the modification of the quark distributions in nuclei, and efforts to improve our understanding of the internal structure of these short-distance and high-momentum configurations in nuclei will provide important input on a wide range of high-energy observables.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Remaining Useful Strength (RUS) Prediction of SiCf-SiCm Composite Materials Using Deep Learning and Acoustic Emission

Prognosis techniques for prediction of remaining useful life (RUL) are of crucial importance to the management of complex systems for they can lead to appropriate maintenance interventions and improvements in reliability. While various data-driven methods have been introduced to predict the remaining useful life (RUL) of machinery systems or batteries, no research has been reported on the remaining useful strength (RUS) prediction of silicon carbide fiber reinforced silicon carbide matrix (SiCf-SiCm) materials with pivotal role in its potential usage as a structural material in nuclear reactors and turbine engines. Knowledge of its degradation process is of the utmost importance to the manufacturers. For this purpose, two approaches based on the machine-learning techniques of random-forest (RF) and convolutional neural network (CNN) are proposed to predict the RUS of SiCf-SiCm using only acoustic emission (AE) signals generated during the material’s stress applying process. Experimental results show that the CNN models achieved better predictive performance than the RF models but the latter with expert-engineered features achieves better prediction for AE signals in the early stage of degradation. Additionally, our results demonstrate that both models can correctly predict the SiCf-SiCm RUS as evaluated by our robust testing method from which the best average root mean square error (RMSE) and Pearson correlation coefficient of 3.55 ksi units and 0.85 were obtained.

36 MATERIALS SCIENCE↗

Monitoring CO 2 Storage at Cranfield, Mississippi with Time-Lapse Offset VSP – Using Integration and Modeling to Reduce Uncertainty

A time-lapse Offset Vertical Seismic Profile (OVSP) data set was acquired as part of a subsurface monitoring program for geologic sequestration of CO 2 . The storage site at Cranfield, near Natchez, Mississippi, is part of a detailed area study (DAS) site for geologic carbon sequestration operated by the U.S. Dept. of Energy’s Southeast Regional Carbon Sequestration Partnership (SECARB). The DAS site includes three boreholes, an injection well and two monitoring wells. The project team selected the DAS site to examine CO 2 sequestration multiphase fluid flow and pressure at the interwell scale in a brine reservoir. The time-lapse (TL) OVSP was part of an integrated monitoring program that included well logs, crosswell seismic, electrical resistance tomography and 4D surface seismic. The goals of the OVSP were to detect the CO 2 induced change in seismic response, give information about the spatial distribution of CO 2 near the injection well and to help tie the high-resolution borehole monitoring to the 4D surface data. The VSP data were acquired in well CFU 31-F1, which is the ~3200 m deep CO 2 injection well at the DAS site. A preinjection survey was recorded in late 2009 with injection beginning in December 2009, and a post injection survey was conducted in Nov 2010 following injection of about 250 kT of CO 2 . The sensor array for both surveys was a 50-level, 3-component, Sercel MaxiWave system with 15 m (49 ft) spacing between levels. The source for both surveys was an accelerated weight drop, with different source trucks used for the two surveys. Consistent time-lapse processing was applied to both data sets. Time-lapse processing generated difference corridor stacks to investigate CO 2 induced reflection amplitude changes from each source point. Corridor stacks were used for amplitude analysis to maximize the signal-to-noise ratio (S/N) for each shot point. Spatial variation in reflectivity (used to ‘map’ the plume) was similar in magnitude to the corridor stacks but, due to relatively lower S/N, the results were less consistent and more sensitive to processing and therefore are not presented. We examined the overall time-lapse repeatability of the OVSP data using three methods, the NRMS and Predictability (Pred) measures of Kragh and Christie (2002) and the signal-to-distortion ratio (SDR) method of Cantillo (2011). Because time-lapse noise was comparable to the observed change, multiple methods were used to analyze data reliability. The reflections from the top and base reservoir were identified on the corridor stacks by correlation with a synthetic response generated from the well logs. A consistent change in the corridor stack amplitudes from pre- to post-CO 2 injection was found for both the top and base reservoir reflections on all ten shot locations analyzed. In addition to the well-log synthetic response, a finite-difference elastic wave propagation model was built based on rock/fluid properties obtained from well logs, with CO 2 induced changes guided by time-lapse crosswell seismic tomography (Ajo-Franklin, et al., 2013) acquired at the DAS site. Time-lapse seismic tomography indicated that two reservoir zones were affected by the flood. The modeling established that interpretation of the VSP trough and peak event amplitudes as reflectivity from the top and bottom of reservoir is appropriate even with possible tuning effects. Importantly, this top/base change gives confidence in an interpretation that these changes arise from within the reservoir, not from bounding lithology. The modeled time-lapse change and the observed field data change from 10 shotpoints are in agreement for both magnitude and polarity of amplitude change for top and base of reservoir. Therefore, we conclude the stored CO 2 has been successfully detected and, furthermore, the observed seismic reflection change can be applied to Cranfield’s 4D surface seismic for spatially delineating the CO 2 /brine interface.

54 ENVIRONMENTAL SCIENCES↗

Controls From Above and Below: Snow, Soil, and Steepness Drive Diverging Trends of Subsurface Water and Streamflow Dynamics

ABSTRACT The importance of subsurface water dynamics, such as water storage and flow partitioning, is well recognised. Yet, our understanding of their drivers and links to streamflow generation has remained elusive, especially in small headwater streams that are often data‐limited but crucial for downstream water quantity and quality. Large‐scale analyses have focused on streamflow characteristics across rivers with varying drainage areas, often overlooking the subsurface water dynamics that shape streamflow behaviour. Here we ask the question: What are the climate and landscape characteristics that regulate subsurface dynamic storage, flow path partitioning, and dynamics of streamflow generation in headwater streams? To answer this question, we used streamflow data and a widely‐used hydrological model (HBV) for 15 headwater catchments across the contiguous United States. Results show that climate characteristics such as aridity and precipitation phase (snow or rain) and land attributes such as topography and soil texture are key drivers of streamflow generation dynamics. In particular, steeper slopes generally promoted more streamflow, regardless of aridity. Streams in flat, rainy sites (< 30% precipitation as snow) with finer soils exhibited flashier regimes than those in snowy sites (> 30% precipitation as snow) or sites with coarse soils and deeper flow paths. In snowy sites, less weathered, thinner soils promoted shallower flow paths such that discharge was more sensitive to changes in storage, but snow dampened streamflow flashiness overall. Results here indicate that land characteristics such as steepness and soil texture modify subsurface water storage and shallow and deep flow partitioning, ultimately regulating streamflow response to climate forcing. As climate change increases uncertainty in water availability, understanding the interacting climate and landscape features that regulate streamflow will be essential to predict hydrological shifts in headwater catchments and improve water resources management.

Kerins, Devon [Department of Civil and Environment↗

Cross-reactive sarbecovirus antibodies induced by mosaic RBD nanoparticles

Broad immune responses are needed to mitigate viral evolution and escape. To induce antibodies against conserved receptor-binding domain (RBD) regions of SARS-like betacoronavirus (sarbecovirus) spike proteins that recognize SARS-CoV-2 variants of concern and zoonotic sarbecoviruses, we developed mosaic-8b RBD nanoparticles presenting eight sarbecovirus RBDs arranged randomly on a 60-mer nanoparticle. Mosaic-8b immunizations protected animals from challenges from viruses whose RBDs were matched or mismatched to those on nanoparticles. Here, we describe neutralizing mAbs isolated from mosaic-8b-immunized rabbits, some on par with Pemgarda, the only currently FDA-approved therapeutic mAb. Deep mutational scanning, in vitro selection of spike resistance mutations, and single-particle cryo-electron microscopy structures of spike–antibody complexes demonstrated targeting of conserved RBD epitopes. Rabbit mAbs included critical D-gene segment RBD-recognizing features in common with human anti-RBD mAbs, despite rabbit genomes lacking an equivalent human D-gene segment, thus demonstrating that the immune systems of humans and other mammals can utilize different antibody gene segments to arrive at similar modes of antigen recognition. These results suggest that animal models can be used to elicit anti-RBD mAbs with similar properties to those raised in humans, which can then be humanized for therapeutic use, and that mosaic RBD nanoparticle immunization coupled with multiplexed screening represents an efficient way to generate and select broadly cross-reactive therapeutic pan-sarbecovirus and pan-SARS-CoV-2 variant mAbs.

Science & Technology - Other Topics↗

Data efficiency and extrapolation trends in neural network interatomic potentials

Abstract Recently, key architectural advances have been proposed for neural network interatomic potentials (NNIPs), such as incorporating message-passing networks, equivariance, or many-body expansion terms. Although modern NNIP models exhibit small differences in test accuracy, this metric is still considered the main target when developing new NNIP architectures. In this work, we show how architectural and optimization choices influence the generalization of NNIPs, revealing trends in molecular dynamics (MD) stability, data efficiency, and loss landscapes. Using the 3BPA dataset, we uncover trends in NNIP errors and robustness to noise, showing these metrics are insufficient to predict MD stability in the high-accuracy regime. With a large-scale study on NequIP, MACE, and their optimizers, we show that our metric of loss entropy predicts out-of-distribution error and data efficiency despite being computed only on the training set. This work provides a deep learning justification for probing extrapolation and can inform the development of next-generation NNIPs.

36 MATERIALS SCIENCE↗

Robust optimization methods for cardiac sparing in tangential breast IMRT

Purpose: In left-sided tangential breast intensity modulated radiation therapy (IMRT), the heart may enter the radiation field and receive excessive radiation while the patient is breathing. The patient’s breathing pattern is often irregular and unpredictable. We verify the clinical applicability of a heart-sparing robust optimization approach for breast IMRT. We compare robust optimized plans with clinical plans at free-breathing and clinical plans at deep inspiration breath-hold (DIBH) using active breathing control (ABC). Methods: Eight patients were included in the study with each patient simulated using 4D-CT. The 4D-CT image acquisition generated ten breathing phase datasets. An average scan was constructed using all the phase datasets. Two of the eight patients were also imaged at breath-hold using ABC. The 4D-CT datasets were used to calculate the accumulated dose for robust optimized and clinical plans based on deformable registration. We generated a set of simulated breathing probability mass functions, which represent the fraction of time patients spend in different breathing phases. The robust optimization method was applied to each patient using a set of dose-influence matrices extracted from the 4D-CT data and a model of the breathing motion uncertainty. The goal of the optimization models was to minimize the dose to the heart while ensuring dose constraints on the target were achieved under breathing motion uncertainty. Results: Robust optimized plans were improved or equivalent to the clinical plans in terms of heart sparing for all patients studied. The robust method reduced the accumulated heart dose (D10cc) by up to 801 cGy compared to the clinical method while also improving the coverage of the accumulated whole breast target volume. On average, the robust method reduced the heart dose (D10cc) by 364 cGy and improved the optBreast dose (D99%) by 477 cGy. In addition, the robust method had smaller deviations from the planned dose to the accumulated dose. The deviation of the accumulated dose from the planned dose for the optBreast (D99%) was 12 cGy for robust versus 445 cGy for clinical. The deviation for the heart (D10cc) was 41 cGy for robust and 320 cGy for clinical. Conclusions: The robust optimization approach can reduce heart dose compared to the clinical method at free-breathing and can potentially reduce the need for breath-hold techniques.

60 APPLIED LIFE SCIENCES↗

Machine Learning for Joint Quality Control

The use of lightweight material combinations has been highly demanded in manufacturing automotive structures. However, making robust dissimilar material joints of such lightweight materials is still challenging. A significant barrier to achieving high-quality and repeatable joint performance is a deficient understanding of the relationship between the welding process, joint attributes, and joint performance. In this context, welding factors refer to material, equipment, environment, and process parameters, while joint features comprise specific microstructural attributes of the weld such as nugget size, heat affected zone (HAZ) topology, intermetallic layer thickness, and sheet thickness reduction. Joint performance is quantified in terms of strength (e.g., tensile shear, coach peel, cross-tension), weld size, and hardness, among other factors. While there have been many attempts to establish this process-structure-property relationship by developing a model derived from the associated physics and first principles, the complexity of the joining processes compounded by the complex interactions with different materials in an automotive assembly line environment, has hindered the usefulness of such attempts. The complexity is further exacerbated using different stacking materials, especially comprising dissimilar material combinations. In practice, the common approach has been the laborious process of creating welds, characterizing them, and then physically testing them through experimentation. With the emergence of artificial intelligence (AI) methods, an alternative pathway to eliciting the desired process-structure-property relationship at an accelerated pace is to use a data-driven approach by employing machine-learning (ML) techniques. This approach is benefitted by the availability of large streams of data, generated through years of research and testing by original equipment manufacturers, in the form of material, process, environmental, equipment, microstructural, and bulk-scale performance information from multimodal, multiscale sensors making measurements from laboratory-scale to production-scale processes. During Phase I efforts, which ended in fiscal year (FY) 2021, the Oak Ridge National Laboratory and Pacific Northwest National Laboratory (ORNL/PNNL) team demonstrated the effectiveness of different ML/AI frameworks in modeling complex relationships between resistance spot welding (RSW) process parameters, weld attributes, and joint properties using a subset of data from General Motors (GM). In FY 2022, the project team further refined and expanded their respective ML models to analyze additional welds with new weld stack-ups and materials to enhance the ML model predictive capability. ORNL extended its unified deep neural networks (DNN) ML training and prediction framework with new data streams of process parameters, and PNNL extended its model describing RSW process parameters’ associations with weld attributes. In FY 2023, the project team completed the development of the AI/ML architecture for analyzing aluminum/steel joints manufactured by GM via RSW and transitioned into the inline welding quality monitoring task for steel/steel RSW joints provided by GM.

36 MATERIALS SCIENCE↗

FALCON: Framework for Anomaly Detection in Industrial Control Systems

Industrial Control Systems (ICS) are used to control physical processes in critical infrastructure. These systems are used in a wide variety of operations such as water treatment, power generation and distribution, and manufacturing. While the safety and security of these systems are of serious concern, recent reports have shown an increase in targeted attacks aimed at manipulating physical processes to cause catastrophic consequences. This trend emphasizes the need for algorithms and tools that provide resilient and smart attack detection mechanisms to protect ICS. In this paper, we propose an anomaly detection framework for ICS based on a deep neural network. The proposed methodology uses dilated convolution and long short-term memory (LSTM) layers to learn temporal as well as long term dependencies within sensor and actuator data in an ICS. The sensor/actuator data are passed through a unique feature engineering pipeline where wavelet transformation is applied to the sensor signals to extract features that are fed into the model. Additionally, this paper explores four variations of supervised deep learning models, as well as an unsupervised support vector machine (SVM) model for this problem. The proposed framework is validated on Secure Water Treatment testbed results. This framework detects more attacks in a shorter period of time than previously published methods.

97 - MATHEMATICS AND COMPUTING↗

Magnetospheric Multiscale (MMS) Mission Commissioning Phase Orbit Determination Error Analysis

The Magnetospheric MultiScale (MMS) mission commissioning phase starts in a 185 km altitude x 12 Earth radii (RE) injection orbit and lasts until the Phase 1 mission orbits and orientation to the Earth-Sun li ne are achieved. During a limited time period in the early part of co mmissioning, five maneuvers are performed to raise the perigee radius to 1.2 R E, with a maneuver every other apogee. The current baseline is for the Goddard Space Flight Center Flight Dynamics Facility to p rovide MMS orbit determination support during the early commissioning phase using all available two-way range and Doppler tracking from bo th the Deep Space Network and Space Network. This paper summarizes th e results from a linear covariance analysis to determine the type and amount of tracking data required to accurately estimate the spacecraf t state, plan each perigee raising maneuver, and support thruster cal ibration during this phase. The primary focus of this study is the na vigation accuracy required to plan the first and the final perigee ra ising maneuvers. Absolute and relative position and velocity error hi stories are generated for all cases and summarized in terms of the ma ximum root-sum-square consider and measurement noise error contributi ons over the definitive and predictive arcs and at discrete times inc luding the maneuver planning and execution times. Details of the meth odology, orbital characteristics, maneuver timeline, error models, and error sensitivities are provided.

Chung, Lauren R.↗

New opportunities at the next-generation neutrino experiments I: BSM neutrino physics and dark matter

The combination of the high intensity proton beam facilities and massive detectors for precision measurements of neutrino oscillation parameters including the charge-parity violating (CPV) phase will open the door to help make beyond the standard model (BSM) physics reachable even in low energy regimes in the accelerator-based experiments. Large-mass detectors with highly precise tracking and energy measurements, excellent timing resolution, and low energy thresholds will enable the searches for BSM phenomena from cosmogenic origin, as well. Therefore, it is also conceivable that BSM topics in the next-generation neutrino experiments could be the dominant physics topics in the foreseeable future, as the precision of the neutrino oscillation parameter and CPV measurements continue to improve. This paper reviews the current landscape of BSM theory in neutrino experiments in two selected areas of the BSM topics-dark matter and neutrino related BSM-and summarizes the current results from existing neutrino experiments to set benchmarks for both theory and experiment. This paper then provides a review of upcoming neutrino experiments throughout the next 10 to 15 year time scale and their capabilities to set the foundation for potential reach in BSM physics in the two aforementioned themes. An important outcome of this paper is to ensure theoretical and simulation tools exist to carry out studies of these new areas of physics, from the first day of the experiments, such as Deep Underground Neutrino Experiment in the U.S. and Hyper-Kamiokande Experiment in Japan.

dark sector↗

Uncertainty guided online ensemble for non-stationary data streams in fusion science

Machine Learning (ML) is poised to play a pivotal role in the development and operation of next-generation fusion devices. Fusion data shows non-stationary behavior with distribution drifts, resulted by both experimental evolution and machine wear-and-tear. ML models assume stationary distribution and fail to maintain performance when encountered with such non-stationary data streams. Online learning techniques have been leveraged in other domains, however it has been largely unexplored for fusion applications. In this paper, we investigate online learning for continuous adaptation to drifting data streams in the prediction of Toroidal Field (TF) coils deflection at the DIII-D fusion facility. We further address the short-term performance degradation inherent to standard online learning, which arises because ground truth is unavailable at prediction time. To mitigate this issue, we propose an uncertainty-guided online ensemble framework. The method leverages the Deep Gaussian Process Approximation (DGPA) for calibrated uncertainty estimation and uses these uncertainty measures to guide a meta-algorithm that aggregates predictions from learners trained over different historical horizons. Our results show that online learning reduces prediction error by 80% compared to a static model. The online ensemble and the proposed uncertainty-guided ensemble further reduce error by approximately 6%, and 10% respectively, relative to standard single-model online learning, while also providing calibrated uncertainty estimates to support operational decision-making.

AI↗

SU-F-19A-11: Retrospective Evaluation of Thermal Coverage by Thermobrachytherapy Seed Arrangements of Clinical LDR Prostate Implants

Purpose: The superiority of a properly-administered combination of radiation therapy and hyperthermia over radiation alone in treatment of human cancers has been demonstrated in multiple studies examining radiobiology, local control, and survival. Unfortunately, hyperthermia is not yet a common modality in oncology practice, due in part to the technical difficulty of heating a deep-seated target volume to sufficient temperature. To address this problem, our group has invented a thermobrachytherapy (TB) seed based on a commonly-used low dose-rate permanent brachytherapy seed for implant in solid tumors. Instead of the tungsten radiographic marker of the standard seed, the TB seed contains one of a self-regulating ferromagnetic alloy. Placement of a patient implanted with such seeds in an oscillating magnetic field generates heat via induction of eddy currents. We present the results of studies of the capability of clinically-realistic TB seed arrangements to adequately heat defined target volumes. Methods: Seed distributions for several past LDR prostate permanent implant brachytherapy patients were reproduced in the finite element analysis software package COMSOL Multiphysics 4.4, with the difference that TB seeds were modelled, rather than the radiation-only seeds actually used for their treatments. The implant geometries were mainly of the modified peripheral loading type; a range of prostatic volumes and blood perfusion rates likely to be seen in a clinical setting were examined. Results: According to the simulations, when distributed to optimize radiation dose, TB seeds also produce sufficient heat to provide thermal coverage of the target given proper selection of the magnetic field strength. However, the thermal distributions may be improved by additional use of hyperthermia-only seeds. Conclusion: A dual-modality seed intended as an alternative to and using the same implantation apparatus and technique as the standard LDR permanent implant seed has been successfully evaluated for its ability to provide sufficient hyperthermia in clinically-realistic implants. This work was partially supported by the National Institutes of Health (NIH) STTR Grant No. R41 CA153631-01A1.

60 APPLIED LIFE SCIENCES↗

Exploring the Use of Novel Spatial Accelerators in Scientific Applications

Driven by the need to find alternative accelerators which can viably replace GPUs in next-generation Supercomputing systems, this paper proposes a methodology to enable agile application/hardware co-design. The application-first methodology provides the ability to come up with design of accelerators while working with real-world workloads, available accelerators, and system software. The iterative design process targets a set of kernels in a workload for performance estimates that can prune the design space for later phases of detailed architectural evaluations. To this effect, in this paper, a novel data-parallel device model is introduced that simulates the latency of performance-sensitive operations in an accelerator including data transfers and kernel computation using multi-core CPUs. The use of off-the-shelf simulators, such as pre-RTL simulator Aladdin or multiple tools available for exploring the design of deep neural network accelerators (e.g., Timeloop) is demonstrated for evaluation of various accelerator designs using applications with realistic inputs. Examples of multiple device configurations that are instantiable in a system are explored to evaluate the performance benefit of deploying novel accelerators. The proposed device is integrated with a programming model and system software to potentially explore the impacts of high-level programming languages/compilers and low-level effects such as task scheduling on multiple accelerators. We analyze our methodology for a set of applications that represent high-performance computing (HPC) and graph analytics. The applications include a computational chemistry kernel realized using tensor contractions, triangle counting, GraphSAGE and Breadth-first Search. These applications include kernels such as dense matrix-dense matrix multiplication, sparse matrix-spare matrix multiplication, and sparse matrix-dense vector multiplication. Our results indicate potential performance benefits and insights for system design by including accelerators that realize these kernels along-side general purpose accelerators.

AI, codesign, Accelerated Computing, Modeling and ↗