Search NASA⌕ Search

SEARCH · Search NASA

Results for “distributed algorithm”

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 1,693 records · Page 94

Extreme Risk Mitigation in Reinforcement Learning using Extreme Value Theory

Risk-sensitive reinforcement learning (RL) has garnered significant attention in recent years due to the growing interest in deploying RL agents in real-world scenarios. A critical aspect of risk awareness involves modelling highly rare risk events (rewards) that could potentially lead to catastrophic outcomes. These infrequent occurrences present a formidable challenge for data-driven methods aiming to capture such risky events accurately. While risk-aware RL techniques do exist, they suffer from high variance estimation due to the inherent data scarcity. Our work proposes to enhance the resilience of RL agents when faced with very rare and risky events by focusing on refining the predictions of the extreme values predicted by the state-action value distribution. To achieve this, we formulate the extreme values of the state-action value function distribution as parameterized distributions, drawing inspiration from the principles of extreme value theory (EVT). We propose an extreme value theory based actor-critic approach, namely, Extreme Valued Actor-Critic (EVAC) which effectively addresses the issue of infrequent occurrence by leveraging EVT-based parameterization. Importantly, we theoretically demonstrate the advantages of employing these parameterized distributions in contrast to other risk-averse algorithms. Our evaluations show that the proposed method outperforms other risk averse RL algorithms on a diverse range of benchmark tasks, each encompassing distinct risk scenarios.

Wang, Yu↗

Poisson Log-Normal Process for Count Data Prediction

Modeling count data is important in physics and other scientific disciplines, where measurements often involve discrete, non-negative quantities such as photon or neutrino detection events. Traditional parametric approaches can be trained to generate integer-count predictions but may struggle with capturing complex, non-linear dependencies often observed in the data. Gaussian process (GP) regression provides a robust non-parametric alternative to modeling continuous data; however, it cannot generate integer outputs. We propose the Poisson Log-Normal (PoLoN) process, a framework that employs GP to model Poisson log-rates. As in GP regression, our approach relies on the correlations between data points captured via GP kernel structure rather than explicit functional parameterizations. We demonstrate that the PoLoN predictive distribution is Poisson-LogNormal and provide an algorithm for optimizing kernel hyperparameters. Furthermore, we adapt the PoLoN approach to the problem of detecting weak localized signals superimposed on a smoothly varying background - a task of considerable interest in many areas of science and engineering. Our framework allows us to predict the strength, location and width of the detected signals. We evaluate PoLoN's performance using both synthetic and real-world datasets, including the open dataset from CERN which was used to detect the Higgs boson at the Large Hadron Collider. Our results indicate that the PoLoN process can be used as a non-parametric alternative for analyzing, predicting, and extracting signals from integer-valued data.

Saha, Anushka [Rutgers U., Piscataway]↗

Large-Scale Multi-Dimensional Document Clustering on GPU Clusters

Document clustering plays an important role in data mining systems. Recently, a flocking-based document clustering algorithm has been proposed to solve the problem through simulation resembling the flocking behavior of birds in nature. This method is superior to other clustering algorithms, including k-means, in the sense that the outcome is not sensitive to the initial state. One limitation of this approach is that the algorithmic complexity is inherently quadratic in the number of documents. As a result, execution time becomes a bottleneck with large number of documents. In this paper, we assess the benefits of exploiting the computational power of Beowulf-like clusters equipped with contemporary Graphics Processing Units (GPUs) as a means to significantly reduce the runtime of flocking-based document clustering. Our framework scales up to over one million documents processed simultaneously in a sixteennode GPU cluster. Results are also compared to a four-node cluster with higher-end GPUs. On these clusters, we observe 30X-50X speedups, which demonstrates the potential of GPU clusters to efficiently solve massive data mining problems. Such speedups combined with the scalability potential and accelerator-based parallelization are unique in the domain of document-based data mining, to the best of our knowledge.

99 GENERAL AND MISCELLANEOUS↗

Use of Dual-wavelength Radar for Snow Parameter Estimates

Use of dual-wavelength radar, with properly chosen wavelengths, will significantly lessen the ambiguities in the retrieval of microphysical properties of hydrometeors. In this paper, a dual-wavelength algorithm is described to estimate the characteristic parameters of the snow size distributions. An analysis of the computational results, made at X and Ka bands (T-39 airborne radar) and at S and X bands (CP-2 ground-based radar), indicates that valid estimates of the median volume diameter of snow particles, D(sub 0), should be possible if one of the two wavelengths of the radar operates in the non-Rayleigh scattering region. However, the accuracy may be affected to some extent if the shape factors of the Gamma function used for describing the particle distribution are chosen far from the true values or if cloud water attenuation is significant. To examine the validity and accuracy of the dual-wavelength radar algorithms, the algorithms are applied to the data taken from the Convective and Precipitation-Electrification Experiment (CaPE) in 1991, in which the dual-wavelength airborne radar was coordinated with in situ aircraft particle observations and ground-based radar measurements. Having carefully co-registered the data obtained from the different platforms, the airborne radar-derived size distributions are then compared with the in-situ measurements and ground-based radar. Good agreement is found for these comparisons despite the uncertainties resulting from mismatches of the sample volumes among the different sensors as well as spatial and temporal offsets.

Liao, Liang↗

On the problem of modeling for parameter identification in distributed structures

Structures are often characterized by parameters, such as mass and stiffness, that are spatially distributed. Parameter identification of distributed structures is subject to many of the difficulties involved in the modeling problem, and the choice of the model can greatly affect the results of the parameter identification process. Analogously to control spillover in the control of distributed-parameter systems, identification spillover is shown to exist as well and its effect is to degrade the parameter estimates. Moreover, as in modeling by the Rayleigh-Ritz method, it is shown that, for a Rayleigh-Ritz type identification algorithm, an inclusion principle exists in the identification of distributed-parameter systems as well, so that the identified natural frequencies approach the actual natural frequencies monotonically from above.

Norris, Mark A.↗

A perturbation technique for parameter identification in distributed structures

Structures are often characterized by parameters, such as mass and stiffness, that are spatially distributed. Parameter identification of distributed structures is subject to many of the difficulties involved in the modeling problem, and the choice of the model can greatly affect the results of the parameter identification process. Analogously to control spillover in the control of distributed-parameter systems, identification spillover is shown to exist as well and its effect is to degrade the parameter estimates. Moreover, as in modeling by the Rayleigh-Ritz method, it is shown that, for a Rayleigh-Ritz type identification algorithm, an inclusion principle exists in the identification of distributed-parameter systems as well, so that the identified natural frequencies approach the actual natural frequencies monotonically from above.

Meirovitch, L.↗

On the problem of modeling for parameter identification in distributed structures

Structures are often characterized by parameters, such as mass and stiffness, that are spatially distributed. Parameter identification of distributed structures is subject to many of the difficulties involved in the modeling problem, and the choice of the model can greatly affect the results of the parameter identification process. Analogously to control spillover in the control of distributed-parameter systems, identification spillover is shown to exist as well and its effect is to degrade the parameter estimates. Moreover, as in modeling by the Rayleigh-Ritz method, it is shown that, for a Rayleigh-Ritz type identification algorithm, an inclusion principle exists in the identification of distributed-parameter systems as well, so that the identified natural frequencies approach the actual natural frequencies monotonically from above.

Norris, Mark A.↗

Dosimetric errors during treatment of centrally located lung tumors with stereotactic body radiation therapy: Monte Carlo evaluation of tissue inhomogeneity corrections

Early experience with stereotactic body radiation therapy (SBRT) of centrally located lung tumors indicated increased rate of high-grade toxicity in the lungs. These clinical results were based on treatment plans that were computed using pencil beam–like algorithms and without tissue inhomogeneity corrections. In this study, we evaluated the dosimetric errors in plans with and without inhomogeneity corrections and with planning target volumes (PTVs) that were within the zone of the proximal bronchial tree (BT). For 10 patients, the PTV, lungs, and sections of the BT either inside or within 2 cm of the PTV were delineated. Two treatment plans were generated for each patient using the following dose-calculation methods: (1) pencil beam (PB) algorithm without inhomogeneity correction (IC) (PB − IC) and (2) PB with inhomogeneity correction (PB + IC). Both plans had identical beam geometry but different beam segment shapes and monitor units (MU) to achieve similar conformal dose coverage of PTV. To obtain the baseline dose distributions, each plan was recalculated using a Monte Carlo (MC) algorithm by keeping MUs the same in the respective plans. The median maximum dose to the proximal BT and PTV dose coverage in the PB + IC plans were overestimated by 8% and 11%, respectively. However, the median maximum dose to the proximal BT and PTV dose coverage in PB − IC plans were underestimated by 15% and 9%. Similar trends were observed in low-dose regions of the lung within the irradiated volume. Our study indicates that dosimetric bias introduced by unit tissue density plans cannot be characterized as underestimation or overestimation of dose without taking the tumor location into account. This issue should be considered when analyzing clinical toxicity data from early lung SBRT trials that utilized unit tissue density for dose calculations.

61 RADIATION PROTECTION AND DOSIMETRY↗

Consensus and Disagreement in Atmospheric River Detection: ARTMIP Global Catalogues

Atmospheric rivers (ARs) constitute an important mechanism for water vapor transport, but research on their characteristics and impacts has relied on a diverse assortment of detection methodologies, complicating comparisons. The AR Tracking Method Intercomparison Project (ARTMIP) provides a platform for comparing such methodologies, but analysis of ARTMIP catalogues has heretofore focused primarily on specific regions. Here we investigate ARs as detected by an ensemble of algorithms with global coverage. We find that the frequency of occurrence of the majority–consensus ARs produces a robust distribution, featuring five hot spots over the extratropical oceans, against which we compare individual algorithm results. We further explore the underlying similarities and differences via two case studies of AR evolution. The dominant source of disagreement between detection methodologies globally consists of detections (or lack thereof) of weak features, and the algorithms otherwise tend to agree remarkably well on the footprints of ARs.

54 ENVIRONMENTAL SCIENCES↗

Face Recognition Algorithm Bias: Performance Differences on Images of Children and Adults

In this work, we examine if current state-of-the-art deep- learning enhanced face recognition systems exhibit a negative bias for children as compared to adults. The systems selected for this work are five top performing1 commercial off-the-shelf (COTS) face recognition systems, two government off-the-shelf (GOTS) face recognition systems and one open-source face recognition solution. The datasets used to evaluate the performance of the systems are both unconstrained in age, pose, illumination, and expression and are publicly available. These datasets are indicative of photo journalistic face datasets published and evaluated on over the last few years. Our findings show a negative bias (i.e., a considerable degradation in performance) for each algorithm on children when compared to the performance obtained on adults. Genuine and imposter distributions high- light the performance bias between the datasets further sup- porting the need for a deeper investigation into algorithm bias as a function of age cohorts. To combat the performance decline on the child demographic, several score- level fusion strategies were evaluated. This work identifies the best score-level fusion technique for child demographic.

Srinivas, Nisha↗

Distributed Grid Control of Flexible Loads and DERs for Optimized Provision of Synthetic Regulating Reserves

Over the course of this project, we have successfully de-risked our distributed microgrid control architecture by tightly integrating its associated control algorithms into a unified software library, installing the software library on several industrial-grade target hardware platforms, and validating the performance of the resulting microgrid controller in a real-life microgrid. Upon completion of the project, we demonstrated that our distributed control architecture is resilient against (i) failures in control devices, (ii) unreliable communication links, (iii) delays in transmitted data, and (iv) imperfect knowledge of the number of (and state of) generation and load assets in the microgrid. In this final report, we present results from all the milestones that were accomplished over the course of this project.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Strategic planning for power system restorations

This paper considers the power system restoration planning problem (PSRPP) for disaster recovery, a fundamental problem faced by all populated areas. PSRPPs are complex stochastic optimization problems that combine resource allocation, warehouse location, and vehicle routing considerations. Furthermore, electrical power systems are complex systems whose behavior can only be determined by physics simulations. Moreover, these problems must be solved under tight runtime constraints to be practical in real-world disaster situations. This work is three fold: (1) it formalizes the specification of PSRPPs; (2) introduces a simple optimization-simulation hybridization necessary for solving PSRPPs; and (3) presents a complete restoration algorithm that utilizes the strengths of mixed integer programming, constraint programming, and large neighborhood search. This paper studied a novel problem in the field of humanitarian logistics, the Power System Restoration Problem (PSRPP). The PSRPP models the strategic planning process for post disaster power system recovery. The paper proposed a multi-stage stochastic hybrid optimization algorithm that yields high quality solutions to real-world benchmarks provided by Los Alamos National Laboratory (LANL). The algorithm uses a variety of technologies, including MIP, constraint programming, and large neighborhood search, to exploit the structure of each individual optimization subproblem. The experimental results on hurricane disaster benchmarks indicate that the algorithm is practical from a computational standpoint and produce significant improvements over existing relief delivery procedures.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Enhancement of Distribution System State Estimation Using Pruned Physics-Aware Neural Networks

Realizing complete observability in the three-phase distribution system remains a challenge that hinders the implementation of classic state estimation algorithms. In this paper, a new method, called the pruned physics-aware neural network (P2N2), is developed to improve the voltage estimation accuracy in the distribution system. The method relies on the physical grid topology, which is used to design the connections between different hidden layers of a neural network model. To verify the proposed method, a numerical simulation based on one- year smart meter data of load consumptions for three-phase power flow is developed to generate the measurement and voltage state data. The IEEE 123-node system is selected as the test network to benchmark the proposed algorithm against the classic weighted least squares (WLS). Numerical results show that P2N2 outperforms WLS in terms of data redundancy and estimation accuracy.

distribution system state estimation↗

Tracking snowmelt during hydrological surface processes using a distributed hydrological model in a mesoscale basin on the Tibetan Plateau

We report that mountain snowpack is an important water resource for the high altitude and latitude regions where the terrain is complex. However, the snowmelt pathway and its actual contribution to streamflow and soil moisture are rarely reported and remain unclear in such regions. To fill in this knowledge gap, we incorporate a snowmelt pathway tracking algorithm to a high-resolution physics-based distributed-hydrology-soil-vegetation model (DHSVM), to track snowmelt movement and quantify snowmelt contributions in the surface hydrologic processes. A simple reservoir operation scheme is also incorporated in the model. The modified model is applied to a dammed meso-scale watershed in the northeast Tibetan Plateau, China to study the snow and reservoir effects. The results show that annual snow contribution to soil moisture (SC-SM) and snow contribution to streamflow (SC-S) significantly decrease over 1965-2019. At a monthly scale, SC-SM has the largest amplitude at the top soil layer and its peak in the deeper layer lags behind the upper layer, and mean monthly SC-S at all stations show bimodal distributions corresponding to snowfall season. Reservoir regulation has subtle impacts (≤2.0%) on SC-S. If the current climate change rate continues, monthly and annual streamflow at the outlet will decrease primarily due to snowpack reduction. To mitigate climate change impacts, better water resource management is needed in this watershed.

54 ENVIRONMENTAL SCIENCES↗

Quantitative evaluation of mask phase defects from through-focus EUV aerial images

Mask defects inspection and imaging is one of the most important issues for any pattern transfer lithography technology. This is especially true for EUV lithography where the wavelength-specific properties of masks and defects necessitate actinic inspection for a faithful prediction of defect printability and repair performance. In this paper we will present a technique to obtain a quantitative characterization of mask phase defects from EUV aerial images. We apply this technique to measure the aerial image phase of native defects on a blank mask, measured with the SEMATECH Berkeley Actinic Inspection Tool (AIT) an EUV zoneplate microscope that operates at Lawrence Berkeley National Laboratory. The measured phase is compared with predictions made from AFM top-surface measurements of those defects. While amplitude defects are usually easy to recognize and quantify with standard inspection techniques like scanning electron microscopy (SEM), defects or structures that have a phase component can be much more challenging to inspect. A phase defect can originate from the substrate or from any level of the multilayer. In both cases its effect on the reflected field is not directly related to the local topography of the mask surface, but depends on the deformation of the multilayer structure. Using the AIT, we have previously showed that EUV inspection provides a faithful and reliable way to predict the appearance of mask defect on the printed wafer; but to obtain a complete characterization of the defect we need to evaluate quantitatively its phase component. While aerial imaging doesn't provide a direct measurement of the phase of the object, this information is encoded in the through focus evolution of the image intensity distribution. Recently we developed a technique that allows us to extract the complex amplitude of EUV mask defects using two aerial images from different focal planes. The method for the phase reconstruction is derived from the Gerchberg-Saxton (GS) algorithm, an iterative method that can be used to reconstruct phase and amplitude of an object from the intensity distributions in the image and in the pupil plane. The GS algorithm is equivalent to a two-parameter optimization problem and it needs exactly two constraints to be solved, namely two intensity distributions in different focal planes. In some formulations, adding any other constraint would result in an ill posed problem. On the other hand, the solution's stability and convergence time can both be improved using more information. We modified our complex amplitude reconstruction algorithm to use an arbitrary number of through focus images and we compared its performance with the previous version in terms of convergence speed, robustness and accuracy. We have demonstrated the phase-reconstruction method on native, mask-blank phase defects and compared the results with phase-predictions made from AFM data collected before and after the multilayer deposition. The method and the current results could be extremely useful for improving the modeling and understanding of native phase defects, their detectability, and their printability.

36 MATERIALS SCIENCE↗

Development of a golden beam data set for the commissioning of a proton double-scattering system in a pencil-beam dose calculation algorithm

Purpose: The purpose of this investigation is to determine if a single set of beam data, described by a minimal set of equations and fitting variables, can be used to commission different installations of a proton double-scattering system in a commercial pencil-beam dose calculation algorithm. Methods: The beam model parameters required to commission the pencil-beam dose calculation algorithm (virtual and effective SAD, effective source size, and pristine-peak energy spread) are determined for a commercial double-scattering system. These parameters are measured in a first room and parameterized as function of proton energy and nozzle settings by fitting four analytical equations to the measured data. The combination of these equations and fitting values constitutes the golden beam data (GBD). To determine the variation in dose delivery between installations, the same dosimetric properties are measured in two additional rooms at the same facility, as well as in a single room at another facility. The difference between the room-specific measurements and the GBD is evaluated against tolerances that guarantee the 3D dose distribution in each of the rooms matches the GBD-based dose distribution within clinically reasonable limits. The pencil-beam treatment-planning algorithm is commissioned with the GBD. The three-dimensional dose distribution in water is evaluated in the four treatment rooms and compared to the treatment-planning calculated dose distribution. Results: The virtual and effective SAD measurements fall between 226 and 257 cm. The effective source size varies between 2.4 and 6.2 cm for the large-field options, and 1.0 and 2.0 cm for the small-field options. The pristine-peak energy spread decreases from 1.05% at the lowest range to 0.6% at the highest. The virtual SAD as well as the effective source size can be accurately described by a linear relationship as function of the inverse of the residual energy. An additional linear correction term as function of RM-step thickness is required for accurate parameterization of the effective SAD. The GBD energy spread is given by a linear function of the exponential of the beam energy. Except for a few outliers, the measured parameters match the GBD within the specified tolerances in all of the four rooms investigated. For a SOBP field with a range of 15 g/cm{sup 2} and an air gap of 25 cm, the maximum difference in the 80%–20% lateral penumbra between the GBD-commissioned treatment-planning system and measurements in any of the four rooms is 0.5 mm. Conclusions: The beam model parameters of the double-scattering system can be parameterized with a limited set of equations and parameters. This GBD closely matches the measured dosimetric properties in four different rooms.

43 PARTICLE ACCELERATORS↗

Modeling the direct sun component in buildings using matrix algebraic approaches: Methods and validation

Simulation tools that enable annual energy performance analysis of optically-complex fenestration systems have been widely adopted by the building industry for use in building design, code development, and the development of rating and certification programs for commercially-available shading and daylighting products. The tools rely on a three-phase matrix operation to compute solar heat gains, using as input low-resolution bidirectional scattering distribution function (BSDF) data (10–15° angular resolution; BSDF data define the angle-dependent behavior of light-scattering materials and systems). Measurement standards and product libraries for BSDF data are undergoing development to support solar heat gain calculations. Simulation of other metrics such as discomfort glare, annual solar exposure, and potentially thermal discomfort, however, require algorithms and BSDF input data that more accurately model the spatial distribution of transmitted and reflected irradiance or illuminance from the sun (0.5° resolution). This study describes such algorithms and input data, then validates the tools (i.e., an interpolation tool for measured BSDF data and the five-phase method) through comparisons with ray-tracing simulations and field monitored data from a full-scale testbed. Simulations of daylight-redirecting films, a micro-louvered screen, and venetian blinds using variable resolution, tensor tree BSDF input data derived from interpolated scanning goniophotometer measurements were shown to agree with field monitored data to within 20% for greater than 75% of the measurement period for illuminance-based performance parameters. The three-phase method delivered significantly less accurate results. We discuss the ramifications of these findings on industry and provide recommendations to increase end user awareness of the current limitations of existing software tools and BSDF product libraries.

14 SOLAR ENERGY↗