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.
Explainable multi-fidelity Bayesian neural network for distribution system state estimation
Distribution System State Estimation (DSSE) is frequently constrained by limited real-time measurements, the uncertainties introduced by distributed energy resources, and the presence of bad data. To address them, this paper proposes an enhanced Multi-Fidelity Bayesian Neural Network (MFBNN) DSSE approach. A low-fidelity layer based on a Deep Neural Network (DNN) is first pre-trained on pseudo-measurement data to learn fundamental state features. Subsequently, a high-fidelity Bayesian Neural Network (BNN) layer leverages limited but high-quality real-time measurements to refine these features, thereby achieving accurate DSSE. Additionally, the deep SHapley Additive exPlanation (SHAP) is developed to quantify the influence of measurement data on DSSE through dual perspectives of global feature importance and local nodal contributions, establishing a hierarchical explainability framework for machine learning-based DSSE. Comparative studies conducted on the IEEE 13-bus system and a real-world 2135-node system from Dominion Energy demonstrate that the proposed method excels in estimation accuracy, even under situations of high noise levels, bad data, and missing data. Further comparisons with Weighted Least Squares (WLS) and other machine learning-based DSSE approaches verify that the proposed framework offers higher accuracy, improved interpretability, and enhanced robustness.
Switching dynamics and ultrafast inversion control of quantum dots for on-chip optical information processing
We demonstrate dynamical near-complete inversion switching of a two-level quantum dot driven by picosecond optical pulses in a bimodal photonic band gap (PBG) waveguide at microwatt power levels. This is enabled by a sharp steplike discontinuity in the local (electromagnetic) density of states (LDOS) provided by a cutoff in one of the two waveguide modes. The atomic Bloch vector equations in this colored vacuum are derived using dressed state description of the atomic density operator in the presence of phonon-mediated dephasing. Instead of assuming phenomenological decay rates, we derive explicit expressions for decay terms of bare state atomic dipole moments and population in the Born-Markov approximation from a master equation approach. Giant Mollow splitting of the atom level due to subwavelength light localization of strong driving pulse causes the time-dependent Mollow bands to straddle the LDOS jump, leading to different radiative decay rates in the upper and lower Mollow sidebands. This results in remarkable field-dependent spontaneous emission and dipolar dephasing rates, combined with a novel 'vacuum structure' term in the Bloch equations. Our Bloch equations predict ultrafast high-contrast inversion switching that is activated and deactivated by picosecond pulse trains detuned below and above the atomic resonance, respectively. This dynamic inversion is due to the rapid rise in relaxation rates as the pulse amplitude rises, causing the Bloch vector to switch from antiparallel to parallel alignment with the pulse 'torque vector'. Subsequent near-complete inversion occurs through an adiabatic following process retained long after the pulse amplitude subsides and the system reverts to slow relaxation. For a 1% inhomogeneously broadened distribution of quantum dot with average of 100 D dipole moment, driven by 1.5 mum picosecond pulses and coupled to a cutoff mode LDOS jump with radiative emission rates of gamma{sub high}=2.5 THz and gamma{sub low}=5 GHz, a large average population switching contrast of 0.5 is demonstrated with a phonon dephasing rate gamma{sub p}=0.5 THz. A 1.6 fJ control pulse is required per switching operation and a 30 muW pulse train is sufficient to maintain the inversion. This switchable gain (loss) segment of the PBG waveguide can be used to controllably amplify (absorb) signal pulses conveying optical information. This provides a robust mechanism for ultrafast multiwavelength channel all-optical logic in PBG microchips.
Deciphering the Scattering of Mechanically Driven Polymers Using Deep Learning
Here, we present a deep learning approach for analyzing two-dimensional scattering data of semiflexible polymers under external forces. In our framework, scattering functions are compressed into a three-dimensional latent space using a Variational Autoencoder (VAE), and two converter networks establish a bidirectional mapping between the polymer parameters (bending modulus, stretching force, and steady shear) and the scattering functions. The training data are generated using off-lattice Monte Carlo simulations to avoid the orientational bias inherent in lattice models, ensuring robust sampling of polymer conformations. The feasibility of this bidirectional mapping is demonstrated by the organized distribution of polymer parameters in the latent space. By integrating the converter networks with the VAE, we obtain a generator that produces scattering functions from given polymer parameters and an inferrer that directly extracts polymer parameters from scattering data. While the generator can be utilized in a traditional least-squares fitting procedure, the inferrer produces comparable results in a single pass and operates 3 orders of magnitude faster. This approach offers a scalable automated tool for polymer scattering analysis and provides a promising foundation for extending the method to other scattering models, experimental validation, and the study of time-dependent scattering data.
Multi-resolution enhancement for full-spectrum neural representations
Scientific data acquisition continues to outpace storage and analysis capabilities, making voxel-basedrepresentations increasingly intractable. Implicit neural representations (INRs) offer a promising solutionby encoding signals through coordinate-based neural networks, serving as surrogates of data, withcomputational and storage requirements scaling with network complexity rather than data dimensionality.However, smaller INRs struggle to faithfully represent multiscale structures, high-frequency informationand fine textures that constitute a large proportion of scientific measurements. We propose WIEN-INR, atheoretically guided hierarchical INR framework that distributes modelling across resolution scales andenables improved representation capacity through a novel enhancement network to recover subtle details.This multiscale architecture allows smaller networks to retain the full spatial-frequency content of thesignal as well as preserve training efficiency and lower storage cost. Evaluated on distinct raw experimentalmeasurements across scales and complexities, WIEN-INR represents a practical step towards a broaderadoption of neural representations in scientific workflows, delivering compact, robust and high-fidelityrepresentations.
FL-DISCO: Federated Generative Adversarial Network for Graph-based Molecule Drug Discovery: Special Session Paper
The outbreak of the global COVID-19 pandemic emphasizes the importance of collaborative drug discovery for high effectiveness; however, due to the stringent data regulation, data privacy becomes an imminent issue needing to be addressed to enable collaborative drug discovery. In addition to the data privacy issue, the efficiency of drug discovery is another key objective since infectious diseases spread exponentially and effectively conducting drug discovery could save lives. Advanced Artificial Intelligence (AI) techniques are promising to solve these problems: (1) Federated Learning (FL) is born to keep data privacy while learning data from distributed clients; (2) graph neural network (GNN) can extract structural properties of molecules whose underlying architecture is the connected atoms; and (3) generative adversarial network (GAN) can generate novel molecules while retaining the properties learned from the training data. In this work, we make the first attempt to build a holistic collaborative and privacy-preserving FL framework, namely FL- DISCO, which integrates GAN and GNN to generate molecular graphs. Experimental results demonstrate the effectiveness of FL- DISCO on: (1) IID data for ESOL and QM9, where FL-DISCO can generate highly novel compounds with high drug-likeliness, uniqueness and LogP scores compared to the baseline; (2) non- IID data for ESOL and QM9, where FL-DISCO generates 100% novel compounds with high validity and LogP scores compared to the baseline. We also demonstrate how different fractions of clients, generator and discriminator architectures affect our evaluation scores.
Automated nuclear cloud feature extraction from film
Chemical, biological, radiological, nuclear, and explosives incidents require rapid detection and characterization for appropriate response. For a nuclear detonation, visible-light cameras may be used to locate the cloud and characterize fallout deposition when coupled with numerical models. Films from the United States’ nuclear testing era compose the only sizeable collection of imagery depicting high-yield detonations. These films offer unique insights into characteristics of flows involving scales that are difficult to replicate experimentally, and they are a valuable source of data for the validation of models for nuclear fallout transport, either as part of emergency response or forensic activities. In this work, we implement modern computer vision and machine learning techniques to identify and track the cloud automatically and subsequently determine the time dependence of some of its features. We trained a ResNet-18 image classifier on hundreds of images to categorize nuclear cloud morphology. Each category or cloud regime is determined by early cloud evolution and is associated to constitutive properties of the flow, such as distribution of vorticity. Next, we identified keypoint features using the KAZE algorithm and tracked these keypoints in the images, allowing us to determine the dimensions and velocities of the cloud across film frames. These measurements converted to real-world units provide valuable experimental data that can be used in the development and validation of nuclear cloud models. We compared the results of this method against manual cloud rise measurements from two different films. In one, our automated method accelerated the feature extraction process without sacrificing measurement accuracy.
The Aerosol Model Benchmarking Repository: A toolkit for model intercomparison
The Aerosol Model Benchmarking Repository and Standards (AMBRS) project was initiated to provide tools and to establish community standards for benchmarking aerosol models. This report describes a set of open-source tools for building, running, and analyzing aerosol box model simulations in a standardized framework. The framework consists of three core components: AMBuilder, a CMake-based build system that compiles supported models consistently; AMBRS, a Python module that defines unified numerical experiments and executes them with aligned inputs; and PyParticle, an aerosol analysis package that standardizes output, computes diagnostics, and visualizes simulation results. Together, these tools enable reproducible intercomparison of aerosol schemes and support process-level evaluation of how model simplifications affect predictions of size distributions, cloud condensation nuclei activity, and other relevant properties relevant for the Earth-Energy system. Beyond its role in benchmarking, AMBRS provides a platform for studying aerosol processes across scales and can be used to generate training data for AI/ML applications in support of a broader hierarchical aerosol modeling strategy.
Rotational Light Curves of Jupiter from Ultraviolet to Mid-infrared and Implications for Brown Dwarfs and Exoplanets
Rotational modulations are observed on brown dwarfs and directly imaged exoplanets, but the underlying mechanism is not well understood. Here we analyze Jupiter’s rotational light curves at 12 wavelengths from the ultraviolet (UV) to the mid-infrared (mid-IR). The peak-to-peak amplitudes of Jupiter’s light curves range from subpercent to 4% at most wavelengths, but the amplitude exceeds 20% at 5 μm, a wavelength sensing Jupiter’s deep troposphere. Jupiter’s rotational modulations are primarily caused by discrete patterns in the cloudless belts instead of the cloudy zones. The light-curve amplitude is controlled by the sizes and brightness contrasts of the Great Red Spot (GRS), expansions of the North Equatorial Belt (NEB), patchy clouds in the North Temperate Belt (NTB), and a train of hot spots in the NEB. In reflection, the contrast is controlled by upper tropospheric and stratospheric hazes, clouds, and chromophores in the clouds. In thermal emission, the small rotational variability is caused by the spatial distribution of temperature and opacities of gas and aerosols; the large variation is caused by the NH{sub 3} cloud holes and thin-thick clouds. The methane-band light curves exhibit opposite-shape behavior compared with the UV and visible wavelengths, caused by a wavelength-dependent brightness change of the GRS. Light-curve evolution is induced by periodic events in the belts and longitudinal drifting of the GRS and patchy clouds in the NTB. This study suggests several interesting mechanisms related to distributions of temperature, gas, hazes, and clouds for understanding the observed rotational modulations on brown dwarfs and exoplanets.
LandScan Global 2023: Silver Edition
For a quarter of a century, the LandScan Global (LSG) project has annually released a global, high-resolution gridded population dataset representing the ambient or unwarned population at a 30 arcsecond resolution. LSG supports a range of applications such as emergency management, disaster response, and human health and security for understanding populations at risk. The 2023 release of LSG, the LandScan Silver Edition, represents a major methodological leap forward while also leveraging previous knowledge—the previous year was the baseline for the current annual update carrying forward valuable knowledge of the built environment for the past quarter century—to train the machine learning models. Compared with annual releases over the past 24years, multiple advancements were made to different aspects of the methodology to achieve reproducibility, transparency, and consistent global propagation of solutions to modeling or population distribution issues identified during the review process. These novel changes include incorporation of the latest available geospatial inputs across the globe, machine learning models instead of manual modifications, population feature importance analysis, open-source solutions vs. proprietary software, generation of multiple global versions, analytic validations, and human-in-the-loop revisions to produce the final version. Additionally, algorithms—such as anomaly detection—were introduced to quickly identify areas of focus to develop a new and robust systematic review. Significant changes in modeled population distributions were observed between the 2022 and 2023 releases, largely attributable to improvements in data and methods and discussed thoroughly within this report. In summation, the LandScan Silver Edition leverages the best of the past quarter century of LSG legacy knowledge and continues a tradition of applying cutting-edge enhancements to serve as a new benchmark for accurate, actionable gridded population data
Electromagnetic shower separation in LArTPCs: study of the calorimetric method at LArIAT
This thesis reports a study of the prospects of using calorimetry for separating photon- from electron-initiated electromagnetic (EM) showers in liquid argon time projection chambers (LArTPCs), a prevalent need for neutrino oscillation experiments based on the LArTPC technology as $\gamma$-initiated showers are a significant background to electron neutrino ($\nu_e$) events, whose signature is electron showers. From events simulated in the LArIAT (Liquid Argon In A Testbeam) experiment's detector, we quantitatively evaluated the dependence of the calorimetric separation on the initial momentum and position (along the beamline) of the primary particle. We could conclude that the separation is more efficient if the primary particle is created inside the chamber, whereas the impact of the momentum is not as significant. Furthermore, a first step was taken towards applying the calorimetric method to the showers measured in LArIAT. A convolutional neural network (CNN) trained on simulation was deployed to find EM showers among the events collected by the experiment. The charge deposition at the beginning of the showers found is centered at a plausible value, although the width of the distribution demands further inquiry.
Probabilistic Inference of Low-Surface-Brightness Galaxy Morphological Parameters Using Simulation-Based Inference
Low-surface-brightness galaxies (LSBGs) are diffuse, often dark-matter-dominated systems whose faintness makes their structural parameters difficult to measure reliably in wide-field imaging surveys. Robust parameter inference, including uncertainty quantification, is important for population studies and for comparisons with models of galaxy formation, as future surveys are expected to produce increasingly large samples of diffuse galaxies. In practice, LSBG profile modeling is sensitive to sky- background errors, masking choices, contaminating background sources, and the computational cost of obtaining posterior-level uncertainties for large samples. Motivated by these questions, we develop a simulation-based inference (SBI) framework for estimating posterior distributions of LSBG morphological parameters from simulated galaxy images. Using PyImfit, we generate DES-like single-Sersic profile LSBG images with known position angle, ellipticity, Sersic index, effective surface brightness, and effective radius. We then train a normalizing-flow-based neural posterior estimator using the sbi package to infer these parameters from the simulated images. For isolated simulated galaxies, the SBI posterior recovers the true input parameters, produces posterior predictive residuals consistent with the assumed noise model, and shows good empirical calibration in a DES-motivated test regime. We also compare SBI with PyImfit-based MCMC inference and find broadly comparable posterior constraints, while SBI enables substantially faster posterior sampling after training. Finally, we test robustness to compact background contaminants. A model trained only on isolated galaxies produces undercovered posteriors on contaminated images, whereas training on simulations with variable contaminant positions and fluxes improves calibration across contaminated test sets. These results demonstrate the promise of SBI for scalable, uncertainty-aware LSBG morphology inference, while emphasizing that posterior reliability strongly depends on whether training simulations include relevant observational complications.
Modeling and Simulation of High-Frequency Solar Irradiance
As the use of solar power as a source of electricity is increasing, so is the interest in modeling radiation at high temporal resolutions. High-dimensional remote sensing data products depend on cloud cover variability, atmosphere aerosol levels, and other atmospheric parameters. Because of weather fronts and aerosols, it is difficult to quantify solar power variability based on distributed solar networks. The global horizontal irradiance (GHI) component in the National Solar Radiation Database (NSRDB) is available at a 30-min time resolution. The algorithm proposed in this paper produces 1-min-resolution GHI samples at locations where NSRDB data are available. The synthetic irradiance datasets are produced using log-additive non-Gaussian mixture models and resampling techniques. The model is trained over historical data, and predicted values are compared with in situ data. This approach allows for estimating the solar irradiance at subhourly temporal resolutions, while featuring variability for locations where measurements are otherwise not available.
Real-time Distribution Simulation and Application Development for Power Systems Education
To help bridge the gap between traditional power system engineering instruction and emerging industry needs, a new set of classroom and research tools are needed. Real-time simulation tools emulating power system control room software present an opportunity to introduce students to the array of operational considerations, technical challenges, and decision-making associated with power system operations. The GridAPPS-D platform is proposed to support coursework and academic research as it provides an open-source platform for simulation, application development, and software integration. The GridAPPS-D platform, simulation capabilities, development environment, and interactive training are discussed in the context of lessons-learned from implementation for undergraduate and graduate students in the US and India. A series of potential GridAPPS-D supported academic capabilities are introduced.
Measurement bias in self-heating x-ray free electron laser experiments from diffraction studies of phase transformation in titanium
X-ray self-heating is a common by-product of X-ray Free Electron Laser (XFEL) techniques that can affect targets, optics, and other irradiated materials. Diagnosis of heating and induced changes in samples may be performed using the x-ray beam itself as a probe. However, the relationship between conditions created by and inferred from x-ray irradiation is unclear and may be highly dependent on the material system under consideration. Here, we report on a simple case study of a titanium foil irradiated, heated, and probed by a MHz XFEL pulse train at 18.1 keV delivered by the European XFEL using measured x-ray diffraction to determine temperature and finite element analysis to interpret the experimental data. We find a complex relationship between apparent temperatures and sample temperature distributions that must be accounted for to adequately interpret the data, including beam averaging effects, multivalued temperatures due to sample phase transitions, and jumps and gaps in the observable temperature near phase transformations. The results have implications for studies employing x-ray probing of systems with large temperature gradients, particularly where these gradients are produced by the beam itself. Finally, this study shows the potential complexity of studying nonlinear sample behavior, such as phase transformations, where biasing effects of temperature gradients can become paramount, precluding clear observation of true transformation conditions.
Feature learning and generalization in deep networks with orthogonal weights
Fully-connected deep neural networks with weights initialized from independent Gaussian distributions can be tuned to criticality, which prevents the exponential growth or decay of signals propagating through the network. However, such networks still exhibit fluctuations that grow linearly with the depth of the network, which may impair the training of networks with width comparable to depth. We show analytically that rectangular networks with tanh activations and weights initialized from the ensemble of orthogonal matrices have corresponding preactivation fluctuations which are independent of depth, to leading order in inverse width. Moreover, we demonstrate numerically that, at initialization, all correlators involving the neural tangent kernel (NTK) and its descendants at leading order in inverse width—which govern the evolution of observables during training—saturate at a depth of ~20, rather than growing without bound as in the case of Gaussian initializations. We speculate that this structure preserves finite-width feature learning while reducing overall noise, thus improving both generalization and training speed in deep networks with depth comparable to width. We provide some experimental justification by relating empirical measurements of the NTK to the superior performance of deep non-linear orthogonal networks trained under full-batch gradient descent on the MNIST and CIFAR-10 classification tasks.
Applying Infrared Thermography as a Method for Online Monitoring of Turbine Blade Coolant Flow
As gas turbine engine manufacturers strive to implement condition-based operation and maintenance, there is a need for blade monitoring strategies capable of early fault detection and root-cause determination. Given the importance of blade cooling flows to turbine blade health and longevity, there is a distinct lack of methodologies for coolant flowrate monitoring. The present study addresses this identified opportunity by applying an infrared thermography system on an engine-representative research turbine to generate data-driven models for prediction of blade coolant flowrate. Thermal images were used as inputs to a linear regression and regularization algorithm to relate blade surface temperature distribution with blade coolant flowrate. Additionally, this study investigates how coolant flowrate prediction accuracy is influenced by the number and breadth of diagnostic measurements. Here, the results of this study indicate that a source of high-fidelity training data can be used to predict blade coolant flowrate within about six percent error. Furthermore, identification of prioritized sensor placement supports application of this technique across multiple sensor technologies capable of measuring blade surface temperature in operating gas turbine engines, including spatially resolved and point-based measurement techniques.
Livermore Big Artificial Neural Network Toolkit
LBANN is a toolkit that is designed to train artificial neural networks efficiently on high performance computing architectures. It is optimized to take advantages of key High Performance Computing features to accelerate neural network training. Specifically it is optimized for low-latency, high bandwidth interconnects, node-local NVRAM, node-local GPU accelerators, and high bandwidth parallel file systems. It is built on top of the open source Elemental distributed-memory dense and spars-direct linear algebra and optimization library that is released under the BSD license. The algorithms contained within LBANN are drawn from the academic literature and implemented to work within a distributed-memory framework.