Search NASA⌕ Search

SEARCH · Search NASA

Results for “Distributed training”

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 73 records · Page 4

DDStore: Distributed Data Store for Scalable Training of Graph Neural Networks on Large Atomistic Modeling Datasets

Graph neural networks (GNNs) are a class of Deep Learning models used in designing atomistic materials for effective screening of large chemical spaces. To ensure robust prediction, GNN models must be trained on large volumes of atomistic data on leadership class supercomputers. Even with the advent of modern architectures that consist of multiple storage layers that include node-local NVMe devices in addition to device memory for caching large datasets, extreme-scale model training faces I/O challenges at scale.We present DDStore, an in-memory distributed data store designed for GNN training on large-scale graph data. DDStore provides a hierarchical, distributed, data caching technique that combines data chunking, replication, low-latency random access, and high throughput communication. DDStore achieves near-linear scaling for training a GNN model using up to 1000 GPUs on the Summit and Perlmutter supercomputers, and reaches up to a 6.15x reduction in GNN training time compared to state-of-the-art methodologies.

Choi, Jong Youl↗

Fair Concurrent Training of Multiple Models in Federated Learning

Federated learning (FL) enables collaborative learning across multiple clients. In most FL work, all clients train a single learning task. However, the recent proliferation of FL applications may increasingly require multiple FL tasks to be trained simultaneously, sharing clients’ computing resources, which we call Multiple-Model Federated Learning (MMFL). Current MMFL algorithms use naïve average-based client-task allocation schemes that often lead to unfair performance when FL tasks have heterogeneous difficulty levels, as the more difficult tasks may need more client participation to train effectively. Furthermore, in the MMFL setting, we face a further challenge that some clients may prefer training specific tasks to others, and may not even be willing to train other tasks, e.g., due to high computational costs, which may exacerbate unfairness in training outcomes across tasks. We address both challenges by firstly designing FedFairMMFL, a difficulty-aware algorithm that dynamically allocates clients to tasks in each training round, based on the tasks’ current performance levels. We provide guarantees on the resulting task fairness and FedFairMMFL’s convergence rate. We then propose novel auction designs that incentivizes clients to train multiple tasks, so as to fairly distribute clients’ training efforts across the tasks, and extend our convergence guarantees to this setting. Here, we finally evaluate our algorithm with multiple sets of learning tasks on real world datasets, showing that our algorithm improves fairness by improving the final model accuracy and convergence speed of the worst performing tasks, while maintaining the average accuracy across tasks.

Federated learning↗

A linear merging methodology for high-resolution precipitation products using spatiotemporal regression

Currently, the only viable option for a global precipitation product is the merger of several precipitation products from different modalities. In this article, we develop a linear merging methodology based on spatiotemporal regression. Four highresolution precipitation products (HRPPs), obtained through methods including the Climate Prediction Center's Morphing (CMORPH), Geostationary Operational Environmental Satellite-Based Auto-Estimator (GOES-AE), GOES-Based Hydro-Estimator (GOES-HE) and Self-Calibrating Multivariate Precipitation Retrieval (SCAMPR) algorithms, are used in this study. The merged data are evaluated against the Arkansas Red Basin River Forecast Center's (ABRFC's) ground-based rainfall product. The evaluation is performed using the Heidke skill score (HSS) for four seasons, from summer 2007 to spring 2008, and for two different rainfall detection thresholds. It is shown that the merged data outperform all the other products in seven out of eight cases. A key innovation of this machine learning method is that only 6% of the validation data are used for the initial training. The sensitivity of the algorithm to location, distribution of training data, selection of input data sets and seasons is also analysed and presented.

Turlapaty, Anish C↗

A Novel Multi-Agent Deep Reinforcement Learning-enabled Distributed Power Allocation Scheme for mmWave Cellular Networks

We consider the power allocation problem over shared spectrum for millimeter-Wave (mmWave) cellular downlink. Existing approaches usually find sub-optimal solutions by solving a non-convex optimization which leads to scalability issues due to centralized control. Therefore, distributed and adaptive approaches are desirable. Recently, model-free Deep Reinforcement Learning (DRL) has achieved success in such wireless resource management tasks. By modeling the radio environment as a Markov Decision Process (MDP) with the base stations (BSs) being the agents, power allocation can be automated at the agent level with comparable throughput performance to conventional centralized schemes. The multi-agent setting presents new challenges as the radio environment is impacted by the joint actions of the agents and is no longer stationary from any individual agent’s perspective. Existing literature bypasses this non-stationarity violation by ignoring it which may cause performance degradation. To tackle this issue, we propose a distributed continuous power allocation scheme based on a modified version of multi-agent Deep Deterministic Policy Gradient (MADDPG) that is tailored for the distributed multiple-agent setting. The proposed scheme employs a centralized-training distributed-execution framework where Q-functions are trained over subsets of BSs while each BS determines its transmit power based only on its own local observation. It admits constant per-BS communication and computation complexity and is thus scalable to large networks. Numerical evaluation shows that the proposed scheme adapts well to a wide range of interference conditions and can achieve comparable or better performance than several state-of-the-art non-learning approaches.

99 GENERAL AND MISCELLANEOUS↗

Estimating luminosities and stellar masses of galaxies photometrically without determining redshifts

Large direct imaging surveys usually use a template-fitting technique to estimate photometric redshifts for galaxies, which are then applied to derive important galaxy properties such as luminosities and stellar masses. These estimates can be noisy and suffer from systematic biases because of the possible mis-selection of templates and the propagation of the photometric redshift uncertainty. We introduce an algorithm, the Direct Empirical Photometric method (DEmP), that can be used to directly estimate these quantities using training sets, bypassing photometric redshift determination. DEmP also applies two techniques to minimize the effects arising from the non-uniform distribution of training set galaxy redshifts from a flux-limited sample. First, for each input galaxy, fitting is performed using a subset of the training set galaxies with photometry and colors closest to those of the input galaxy. Second, the training set is artificially resampled to produce a flat distribution in redshift or other properties, e.g., luminosity. To test the performance of DEmP, we use a four filter-band mock catalog to examine its ability to recover redshift, luminosity, stellar mass, and luminosity and stellar mass functions. We also compare the results to those from two publicly available template-fitting methods, finding that the DEmP algorithm outperforms both. We find that resampling the training set to have a uniform redshift distribution produces the best results not only in photometric redshift, but also in estimating luminosity and stellar mass. The DEmP method is especially powerful in estimating quantities such as near-IR luminosities and stellar mass using only data from a small number of optical bands.

79 ASTRONOMY AND ASTROPHYSICS↗

An Examination of Diameter Density Prediction with k-NN and Airborne Lidar

While lidar-based forest inventory methods have been widely demonstrated, performances of methods to predict tree diameters with airborne lidar (lidar) are not well understood. One cause for this is that the performance metrics typically used in studies for prediction of diameters can be difficult to interpret, and may not support comparative inferences between sampling designs and study areas. To help with this problem we propose two indices and use them to evaluate a variety of lidar and k nearest neighbor (k-NN) strategies for prediction of tree diameter distributions. The indices are based on the coefficient of determination ( R 2 ), and root mean square deviation (RMSD). Both of the indices are highly interpretable, and the RMSD-based index facilitates comparisons with alternative (non-lidar) inventory strategies, and with projects in other regions. K-NN diameter distribution prediction strategies were examined using auxiliary lidar for 190 training plots distribute across the 800 km 2 Savannah River Site in South Carolina, USA. In conclusion, we evaluate the performance of k-NN with respect to distance metrics, number of neighbors, predictor sets, and response sets. K-NN and lidar explained 80% of variability in diameters, and Mahalanobis distance with k = 3 neighbors performed best according to a number of criteria.

60 APPLIED LIFE SCIENCES↗

Hierarchical Inference with Bayesian Neural Networks: An Application to Strong Gravitational Lensing

In the past few years, approximate Bayesian Neural Networks (BNNs) have demonstrated the ability to produce statistically consistent posteriors on a wide range of inference problems at unprecedented speed and scale. However, any disconnect between training sets and the distribution of real-world objects can introduce bias when BNNs are applied to data. This is a common challenge in astrophysics and cosmology, where the unknown distribution of objects in our universe is often the science goal. In this work, we incorporate BNNs with flexible posterior parameterizations into a hierarchical inference framework that allows for the reconstruction of population hyperparameters and removes the bias introduced by the training distribution. We focus on the challenge of producing posterior PDFs for strong gravitational lens mass model parameters given Hubble Space Telescope–quality single-filter, lens-subtracted, synthetic imaging data. We show that the posterior PDFs are sufficiently accurate (statistically consistent with the truth) across a wide variety of power-law elliptical lens mass distributions. We then apply our approach to test data sets whose lens parameters are drawn from distributions that are drastically different from the training set. We show that our hierarchical inference framework mitigates the bias introduced by an unrepresentative training set's interim prior. Simultaneously, we can precisely reconstruct the population hyperparameters governing our test distributions. Our full pipeline, from training to hierarchical inference on thousands of lenses, can be run in a day. The framework presented here will allow us to efficiently exploit the full constraining power of future ground- and space-based surveys (https://github.com/swagnercarena/ovejero).

79 ASTRONOMY AND ASTROPHYSICS↗

Improving Enzyme Optimum Temperature Prediction with Resampling Strategies and Ensemble Learning

Accurate prediction of the optimal catalytic temperature ( T opt ) of enzymes is vital in biotechnology, as enzymes with high T opt values are desired for enhanced reaction rates. Recently, a machine learning method (temperature optima for microorganisms and enzymes, TOME) for predicting T opt was developed. TOME was trained on a normally distributed data set with a median T opt of 37 °C and less than 5% of T opt values above 85 °C, limiting the method’s predictive capabilities for thermostable enzymes. Due to the distribution of the training data, the mean squared error on T opt values greater than 85 °C is nearly an order of magnitude higher than the error on values between 30 and 50 °C. Here, we apply ensemble learning and resampling strategies that tackle the data imbalance to significantly decrease the error on high T opt values (>85 °C) by 60% and increase the overall R 2 value from 0.527 to 0.632. The revised method, temperature optima for enzymes with resampling (TOMER), and the resampling strategies applied in this work are freely available to other researchers as Python packages on GitHub.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Distributed Stochastic Optimization of a Neural Representation Network for Time-Space Tomography Reconstruction

4D time-space reconstruction of dynamic events or deforming objects using X-ray computed tomography (CT) is an important inverse problem in non-destructive evaluation. Conventional back-projection based reconstruction methods assume that the object remains static for the duration of several tens or hundreds of X-ray projection measurement images (reconstruction of consecutive limited-angle CT scans). However, this is an unrealistic assumption for many in-situ experiments that causes spurious artifacts and inaccurate morphological reconstructions of the object. To solve this problem, we propose to perform a 4D time-space reconstruction using a distributed implicit neural representation (DINR) network that is trained using a novel distributed stochastic training algorithm. Our DINR network learns to reconstruct the object at its output by iterative optimization of its network parameters such that the measured projection images best match the output of the CT forward measurement model. Here, we use a forward measurement model that is a function of the DINR outputs at a sparsely sampled set of continuous valued 4D object coordinates. Unlike previous neural representation architectures that forward and back propagate through dense voxel grids that sample the object's entire time-space coordinates, we only propagate through the DINR at a small subset of object coordinates in each iteration resulting in an order-of-magnitude reduction in memory and compute for training. DINR leverages distributed computation across several compute nodes and GPUs to produce high-fidelity 4D time-space reconstructions. We use both simulated parallel-beam and experimental cone-beam X-ray CT datasets to demonstrate the superior performance of our approach.

36 MATERIALS SCIENCE↗

Probability of Pinching Failure in PWR SNF after Long-Term Storage

A probabilistic assessment was performed on pinching failure of PWR SNF rod cladding under accidental SNF cask transport conditions. a set of 3,000 rod pinching scenarios was analyzed, with different geometrical, material, and loading parameters, using a Latin Hypercube Sampling (LHS). The investigated accidents include train derailments leading to cask collisions or cask drops. Pinching failure was related to cladding offset strain, defined as the plastic deformation interval in radial cladding direction normalized with respect to the outer cladding diameter. Finite element (FE) models were used to compute SNF rod deformations and the offset strain demand OSD in each of the assessed pinching scenarios. The numerical models account for the contribution of the degraded fuel pellets to rod pinching stiffness. The offset strain capacity OSC was computed using regression and interpolation models based on ring compression (RCT) test results, and account for hydride-related effects on cladding ductility. The results show a low probability of pinching failure. Hydride-related cladding embrittlement affects the structural integrity of PWR SNF rods after long-term dry storage of up to 300 years. Radial hydride patterns or large amounts of circumferentially oriented hydrides reduce the cladding ductile deformation capability when pinched at low material temperatures. The probability of hydride-related pinching failure of PWR cladding is estimated after long-term dry storage to address: How do PWR SNF rod conditions like hydrogen content, vacuum drying peak cladding hoop stress σ{sub θ}, and cladding material temperature affect cladding ductility? Is hydride-related cladding embrittlement a material degradation mechanism that affects the structural integrity of PWR SNF rods under pinching loading? What are the controlling parameters for deformation demands in PWR SNF rods under pinching loading? Is the pinching load amplitude an important parameter that controls the probability of pinching failure? How do parameters like pellet-cladding gap or fuel pellet degradation affect the pinching loading capacity of PWR SNF rods? Statistical modelling of PWR SNF rod conditions in U.S. inventory: A pseudo rod database was generated based on EIA database. Analysis of train accident scenarios: a probability density function (pdf) of a pinching load distribution P{sub 9m} of a 9-m cask drop is linearly scaled to generate a more representative: distribution P{sub AS}. The new distribution considers train collision velocities, cask drops for on- and off-bridge accidents, and cask impact orientation. LHS generates 3,000 rod pinching scenarios based on nine independent geometrical, material, and loading parameters. Probabilistic FE Analysis: The LHS realizations are used to generate 3,000 FE models. The FE analysis is conducted for P{sub 9m}, and P{sub AS}. The workflow is automatized. The OSC is computed with regression and interpolation models, as a function of the hydrogen content C{sub H}, vacuum drying peak σ{sub θ}, and SNF temperature T{sub FR}: The regression data was obtained in RCTs on PWR cladding. The OSD is derived from the cladding deformations in the FE models. Cladding failure was identified via a comparison of OSC and OSD. A 300 and a 30 years-of-dry-storage duration is investigated: The dry-storage period affects the expected SNF temperature and consequently, the OSC and the OSD. The material temperature LHS population is adapted to account for a SNF temperature variation. The probability of hydride-related, ductile cladding failure during a severe transportation accident is not significantly affected by the pinching load amplitude: The probability is slightly reduced from a range of less than 0.0333% to 0.1667% for P{sub 9m}, to a range of less than 0.0333% to 0.1333% for P{sub AS}, depending on the vacuum drying peak cladding hoop stress modelling assumptions. Further, the SNF dry-storage duration does not significantly affect the probability of ductile pinching failure. Based on available train accidents data, the pinching failure probability for a 5,000 km long transportation route ranges from less than 1.62 E-7 to a value of 6.47 E-7, depending on the vacuum drying peak cladding hoop stress modelling assumptions. The pinching load amplitude is not a controlling parameter for the probability of ductile cladding pinching failure, and does not control the cladding deformations once pellet-cladding contact is established, The pellet-cladding gap width, however, plays a key role. Fuel pellet degradation only affects the deformation capacity, if the pellet stiffness is significantly reduced (more than 90%). Only highly embrittled cladding is likely to fail. The deformation demands on the cladding under pinching loading are small.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

USAID Colombia Young Leaders Workforce Training Program Action Plans: Forecasting Distributed Photovoltaic Adoption in Barranquilla, Colombia

As part of the U.S. Agency for International Development (USAID)-National Renewable Energy Laboratory (NREL) Young Leaders Workforce Training Program in Colombia, the Association of Renewable Energies Colombia (SER) participants leveraged their training and professional experience to develop an action plan for modeling the projected adoption of distributed solar photovoltaics (PV) out to 2050 for the city of Barranquilla, Colombia. This case study provides an overview of the key activities and outcomes of the Distributed Generation Market Demand Model (dGen™) Colombia project.

14 SOLAR ENERGY↗

Optimized asynchronous training of neural networks using a distributed parameter server with eager updates

A method of training a neural network includes, at a local computing node, receiving remote parameters from a set of one or more remote computing nodes, initiating execution of a forward pass in a local neural network in the local computing node to determine a final output based on the remote parameters, initiating execution of a backward pass in the local neural network to determine updated parameters for the local neural network, and prior to completion of the backward pass, transmitting a subset of the updated parameters to the set of remote computing nodes.

Hamidouche, Khaled↗

Machine Learning for Distributed Acoustic Sensing data (MLDAS) v1.0.1

MLDAS is a Python-written package for exploratory data analysis and deep learning training on Distributed Acoustic Sensing data. The machine learning tools are powered by the PyTorch library and designed to work efficiently on large scale datasets using parallel computing. Various SLURM scripts as well as a tutorial have also been made available to allow geophysicists to quickly and easily implement the available tools in their analysis workflow on supercomputer facilities.

Dumont, Vincent↗

Adversarial sampling of unknown and high-dimensional conditional distributions

Many engineering problems require the prediction of realization-to-realization variability or a refined description of modeled quantities. In that case, it is necessary to sample elements from unknown high-dimensional spaces with possibly millions of degrees of freedom. While there exist methods able to sample elements from probability density functions (PDF) with known shapes, several approximations need to be made when the distribution is unknown. In this paper the sampling method, as well as the inference of the underlying distribution, are both handled with a data-driven method known as generative adversarial networks (GAN), which trains two competing neural networks to produce a network that can effectively generate samples from the training set distribution. In practice, it is often necessary to draw samples from conditional distributions. When the conditional variables are continuous, only one (if any) data point corresponding to a particular value of a conditioning variable may be available, which is not sufficient to estimate the conditional distribution. This work handles this problem using an a priori estimation of the conditional moments of a PDF. Herein, two approaches, stochastic estimation, and an external neural network are compared for computing these moments; however, any preferred method can be used. The algorithm is demonstrated in the case of the deconvolution of a filtered turbulent flow field. It is shown that all the versions of the proposed algorithm effectively sample the target conditional distribution with minimal impact on the quality of the samples compared to state-of-the-art methods. Additionally, the procedure can be used as a metric for the diversity of samples generated by a conditional GAN (cGAN) conditioned with continuous variables.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Roadmap and Benchmarking: Privacy in Federated Load Forecasting

Data-driven techniques for energy demand forecasting continue to emerge with promising impacts on distribution grid planning. However, the development of robust and generalizable machine learning models requires that representative high quality training data are available. Distributed energy resources have begun to embed intelligence, gathering large amounts of data on customer demand, behavior, and household devices that are connected to the grid. Though utilities aggregate meter-level demand data for load shaping, demand response, outage management, reliability planning, and billing applications, there lies an inherent privacy concern in sharing consumption data that may identify individual consumer behavioral patterns. Hence, while sharing the data is crucial, the private sensitive customer data must be safeguarded from being exposed or manipulated. In this study, we propose a roadmap for implementing a based privacy preserving framework to support the advancement of data-driven analytics in data-sensitive distributed energy resources environments. The roadmap incorporates federated learning–a distributed training framework, differential privacy–a statistical framework that provides guarantees to safeguard the leakage of sensitive data, secure multiparty computation and homomorphic encryption– techniques for encrypting model gradients and applying secure aggregation on the server. Moreover, we perform baseline experiments on the federated short-term load forecasting (STLF) task using open-source residential load profile datasets, offering insights into the challenges of integrating differential privacy into federated learning.

Abebe, Waqwoya [Oak Ridge National Laboratory (ORN↗

DNTTD (Distributed Non-Negative Tensor Train Decomposition)

The era of exascale computing opens new venues for innovations and discoveries in many scientific, engineering, and commercial fields. However, with the exa flops also come the extra-large high-dimensional data generated by high performance computing. High-dimensional data is presented as multidimensional arrays, aka tensors. The presence of latent (not directly observable) structures in the tensor allows a unique representation and compression of the data by classical tensor factorization techniques. However, the classical tensor methods are not always stable or they can be exponential in their memory requirements, which makes them not suitable for high-dimensional tensors. Tensor train (TT) is a state-of-the-art tensor network introduced for factorization of high-dimensional tensors. TT transforms the initial high-dimensional tensor in a network of three-dimensional tensors that requires only a linear storage. Many real-world data, such as, density, temperature, population, probability, etc., are non-negative and for an easy interpretation, the algorithms preserving non-negativity are preferred. Here, we introduce a distributed non-negative tensor-train and demonstrate its scalability and the compression on synthetic and real world big datasets.

Bhattarai, Manish↗

DP-TwoLevel: two-stage gradient subspace learning for differentially private federated learning

Federated learning (FL) enables collaborative model training across distributed data sources without sharing raw data, but faces fundamental challenges in communication efficiency and privacy. Differentially private (DP) training mitigates information leakage but introduces noise that degrades model performance, especially in high-dimensional settings. We propose DP-TwoLevel, a hierarchical gradient projection method that improves utility under fixed DP constraints by exploiting low-dimensional structure in model updates. Our approach learns a two-level PCA-based representation of gradients and applies DP noise in a reduced-dimensional subspace, thereby lowering the effective noise magnitude while preserving dominant signal components. We evaluate the method across three datasets (MNIST, Fashion-MNIST, CIFAR-10) and three privacy regimes (ϵ∈0.5, 1.0, 2.0). Across nine experimental settings, DP-TwoLevel consistently outperforms DP-FedAvg, achieving an average accuracy improvement of 9.44%, with larger gains observed in lower ϵ(higher-noise) regimes (up to +22.31%). We further analyze scalability across models ranging from 100K to 1.49M parameters and identify a variance-based success criterion: performance remains strong when the projection preserves more than 75% of gradient variance, degrades in a marginal regime (65–75%), and fails below this threshold. Our results demonstrate that structure-aware dimensionality reduction can significantly improve the privacy–utility tradeoff in FL without modifying formal privacy guarantees. We also provide empirical evidence of scaling limitations for global projections and motivate per-layer extensions for larger models.

Kotevska, Olivera [ORNL] (ORCID:0000000316772243)↗

Grid Engineering for Accelerated Renewable Energy Deployment (GEARED): National Network Administrator (Final Report)

DOE developed the five-year, Grid Engineering for Accelerated Renewable Energy Deployment (GEARED) initiative and provided funding to create four Distributed Technology Training Consortia (DTTC) made up of universities, utilities, and organizations. These DTTCs supported the inclusion of power systems analysis, research and development, into training activities such as curriculum and short course development, internships & coops, and continuing education. Additionally, DOE aimed to build a national framework for power systems training and curriculum that accelerates the growth of power systems programs and human capacity. To that end, the GEARED initiative utilized a National Network Administrator (NNA) to manage and leverage the activities of the four DTTCs as well as that of others participating in distributed technology and engineering through research or practice. DOE chose the Interstate Renewable Energy Council (IREC) as the NNA.

14 SOLAR ENERGY↗