Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “heterogeneous networks”

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 613 records · Page 34
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Magnetic resonance insights into the heterogeneous, fractal-like kinetics of chemically recyclable polymers

Moving toward a circular plastics economy is a vital aspect of global resource management. Chemical recycling of plastics ensures that high-value monomers can be recovered from depolymerized plastic waste, thus enabling circular manufacturing. However, to increase chemical recycling throughput in materials recovery facilities, the present understanding of polymer transport, diffusion, swelling, and heterogeneous deconstruction kinetics must be systematized to allow industrial-scale process design, spanning molecular to macroscopic regimes. To develop a framework for designing depolymerization processes, we examined acidolysis of circular polydiketoenamine elastomers. We used magnetic resonance to monitor spatially resolved observables in situ and then evaluated these data with a fractal method that treats nonlinear depolymerization kinetics. This approach delineated the roles played by network architecture and reaction medium on depolymerization outcomes, yielding parameters that facilitate comparisons between bulk processes. These streamlined methods to investigate polymer hydrolysis kinetics portend a general strategy for implementing chemical recycling on an industrial scale.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2021

INTEGRATE - Inverse Network Transformations for Efficient Generation of Robust Airfoil and Turbine Enhancements

The INTEGRATE (Inverse Network Transformations for Efficient Generation of Robust Airfoil and Turbine Enhancements) project is developing a new inverse-design capability for the aerodynamic design of wind turbine rotors using invertible neural networks. This AI-based design technology can capture complex non-linear aerodynamic effects while being 100 times faster than design approaches based on computational fluid dynamics. This project enables innovation in wind turbine design by accelerating time to market through higher-accuracy early design iterations to reduce the levelized cost of energy. INVERTIBLE NEURAL NETWORKS Researchers are leveraging a specialized invertible neural network (INN) architecture along with the novel dimension-reduction methods and airfoil/blade shape representations developed by collaborators at the National Institute of Standards and Technology (NIST) learns complex relationships between airfoil or blade shapes and their associated aerodynamic and structural properties. This INN architecture will accelerate designs by providing a cost-effective alternative to current industrial aerodynamic design processes, including: - Blade element momentum (BEM) theory models: limited effectiveness for design of offshore rotors with large, flexible blades where nonlinear aerodynamic effects dominate - Direct design using computational fluid dynamics (CFD): cost-prohibitive - Inverse-design models based on deep neural networks (DNNs): attractive alternative to CFD for 2D design problems, but quickly overwhelmed by the increased number of design variables in 3D problems AUTOMATED COMPUTATIONAL FLUID DYNAMICS FOR TRAINING DATA GENERATION - MERCURY FRAMEWORK The INN is trained on data obtained using the University of Marylands (UMD) Mercury Framework, which has with robust automated mesh generation capabilities and advanced turbulence and transition models validated for wind energy applications. Mercury is a multi-mesh paradigm, heterogeneous CPU-GPU framework. The framework incorporates three flow solvers at UMD, 1) OverTURNS, a structured solver on CPUs, 2) HAMSTR, a line based unstructured solver on CPUs, and 3) GARFIELD, a structured solver on GPUs. The framework is based on Python, that is often used to wrap C or Fortran codes for interoperability with other solvers. Communication between multiple solvers is accomplished with a Topology Independent Overset Grid Assembler (TIOGA). NOVEL AIRFOIL SHAPE REPRESENTATIONS USING GRASSMAN SPACES We developed a novel representation of shapes which decouples affine-style deformations from a rich set of data-driven deformations over a submanifold of the Grassmannian. The Grassmannian representation as an analytic generative model, informed by a database of physically relevant airfoils, offers (i) a rich set of novel 2D airfoil deformations not previously captured in the data , (ii) improved low-dimensional parameter domain for inferential statistics informing design/manufacturing, and (iii) consistent 3D blade representation and perturbation over a sequence of nominal shapes. TECHNOLOGY TRANSFER DEMONSTRATION - COUPLING WITH NREL WISDEM Researchers have integrated the inverse-design tool for 2D airfoils (INN-Airfoil) into WISDEM (Wind Plant Integrated Systems Design and Engineering Model), a multidisciplinary design and optimization framework for assessing the cost of energy, as part of tech-transfer demonstration. The integration of INN-Airfoil into WISDEM allows for the design of airfoils along with the blades that meet the dynamic design constraints on cost of energy, annual energy production, and the capital costs. Through preliminary studies, researchers have shown that the coupled INN-Airfoil + WISDEM approach reduces the cost of energy by around 1% compared to the conventional design approach. This page will serve as a place to easily access all the publications from this work and the repositories for the software developed and released through this pr...

aerodynamics↗
NASA NTRS2021

Integrating continuous atmospheric boundary layer and tower-based flux measurements to advance understanding of land-atmosphere interactions

The atmospheric boundary layer mediates the exchange of energy, matter, and momentum between the land surface and the free troposphere, integrating a range of physical, chemical, and biological processes and is defined as the lowest layer of the atmosphere (ranging from a few meters to 3 km). In this review, we investigate how continuous, automated observations of the atmospheric boundary layer can enhance the scientific value of co-located eddy covariance measurements of land-atmosphere fluxes of carbon, water, and energy, as are being made at FLUXNET sites worldwide. We highlight four key opportunities to integrate tower-based flux measurements with continuous, long-term atmospheric boundary layer measurements: (1) to interpret surface flux and atmospheric boundary layer exchange dynamics and feedbacks at flux tower sites, (2) to support flux footprint modelling, the interpretation of surface fluxes in heterogeneous and mountainous terrain, and quality control of eddy covariance flux measurements, (3) to support regional-scale modeling and upscaling of surface fluxes to continental scales, and (4) to quantify land-atmosphere coupling and validate its representation in Earth system models. Adding a suite of atmospheric boundary layer measurements to eddy covariance flux tower sites, and supporting the sharing of these data to tower networks, would allow the Earth science community to address new emerging research questions, better interpret ongoing flux tower measurements, and would present novel opportunities for collaborations between FLUXNET scientists and atmospheric and remote sensing scientists.

Manuel Helbig↗
DOE OSTI2026

Multi-scale structural analysis of swift heavy ion-irradiated ZrO 2 and HfO 2

The radiation-induced monoclinic-to-tetragonal phase transition in ZrO 2 and HfO 2 has been the subject of many investigations, but the transformation pathways and underlying structural mechanisms are still not well understood. In this study, microcrystalline powder samples of ZrO 2 and HfO 2 were irradiated with 946 MeV and 1470 MeV Au ions to a wide fluence range up to 3 × 10 13 ions/cm 2 . To characterize beam-induced structural modifications across all spatial length scales, complementary experimental techniques such as synchrotron X-ray diffraction and spallation neutron total scattering were used. The phase evolution of the tetragonal polymorph with increasing ion fluence is accurately described by a heterogeneous track-overlap model that incorporates both direct- and double-impact processes. These damage accumulation processes are an expression of a core-shell ion track morphology that depends on irradiation conditions and target material. Neutron pair distribution function analysis revealed that ion-beam-induced tetragonal ZrO 2 is merely a configurational average of short-range orthorhombic (Pbcn) domains stabilized by a dense network of domain walls. Furthermore, this knowledge is critical for a better understanding of how crystalline-to-crystalline phase transformations proceed at the atomic scale under extreme conditions.

Monoclinic-to-tetragonal phase transformation↗
DOE OSTI2025

Multi-frequency progressive refinement for learned inverse scattering

Interpreting scattered acoustic and electromagnetic wave patterns is a computational task that enables remote imaging in a number of important applications, including medical imaging, geophysical exploration, sonar and radar detection, and nondestructive testing of materials. However, accurately and stably recovering an inhomogeneous medium from far-field scattered wave measurements is a computationally difficult problem, due to the nonlinear and non-local nature of the forward scattering process. We design a neural network, called Multi-Frequency Inverse Scattering Network (MFISNet), and a training method to approximate the inverse map from far-field scattered wave measurements at multiple frequencies. We consider three variants of MFISNet, with the strongest performing variant inspired by the recursive linearization method — a commonly used technique for stably inverting scattered wavefield data — that progressively refines the estimate with higher frequency content. MFISNet outperforms past methods in regimes with high-contrast, heterogeneous large objects, and inhomogeneous unknown backgrounds.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Automated Generation of Integrated Digital and Spiking Neuromorphic Machine Learning Accelerators

The growing numbers of application areas for artificial intelligence (AI) methods have led to an explosion of domain-specific accelerators that could support every new machine learning (ML) algorithm advancement, clearly highlighting the need for a capability to quickly and automatically transition from algorithm definition to hardware implementation and explore design space along a variety of SWaP (size, weight and Power). The software defined architectures (SODA) synthesizer implements a compiler-based modular infrastructure for the end-to-end generation of machine learning accelerators from high-level frameworks to hardware description language. At the same time, neuromorphic computing, by mimicking how the brain operates, promises to perform artificial intelligence tasks at efficiencies orders of magnitude higher than the current conventional tensor-processing based accelerators, as demonstrated by a variety of specialized designs leveraging Spiking Neural Networks (SNNs). Nevertheless, the mapping of an artificial neural network (ANN) to solutions supporting SNNs is still a non-trivial and very device-specific task, and completely lack the possibility to design hybrid systems that integrate conventional and spiking neural models. In this paper we discuss the support for such an integrated generation leveraging the SODA Synthesizer framework and its modular structure. In particular, we present a new MLIR dialect (part of the SODA frontend) that allows expressing spiking neural network features (e.g., available resources, spiking sequences, analog signal reading, etc.) and illustrate how it enables mapping to Spiking Neurons and deployment to the related specialized hardware (which, in the digital domain, could be generated through the other existing layers of the SODA Synthesizer). We then discuss the opportunities for even deeper integration afforded by the hardware compilation infrastructure, providing a path towards the generation of complex heterogeneous artificial intelligence systems.

Curzel, Serena↗
DOE OSTI2025

Improved Gas Plume Identification Using Nearest Neighbor Methods for Background Estimation

Longwave infrared (LWIR) hyperspectral imaging (HSI) can be used for many tasks in remote sensing, including detecting and identifying effluent gases by LWIR sensors on airborne platforms. Identification is used after detection to increase confidence in weakly detected plumes, reduce false positives from detection, and distinguish between similar and confounding material signatures. Background estimation is an important step used to reveal the unique spectral characteristics of the detected gas, allowing the identification model to determine what the gas is specifically. The importance of proper background estimation increases when dealing with weak signals, large libraries of gases of interest, and uncommon or heterogeneous backgrounds. In this article, we propose two methods for background estimation: a novel k-nearest segments (KNS) algorithm and the standard k-nearest neighbors (KNN) algorithm. We test our methods and three existing background estimation methods for comparison against global background estimation to determine which performs best at estimating the true background radiance under a plume and for increasing identification confidence using a neural network classification model. We compare the different methods using 640 simulated weak plumes in an urban environment. For identification, our KNS algorithm improves median neural network identification confidence by 53.2%. For background radiance estimation, the KNN algorithm provides a median of 49 times less RMSE than global background estimation. Furthermore, KNN is the easiest method to tune for different plumes, making it an excellent “out of the box” background estimator.

47 OTHER INSTRUMENTATION↗
DOE OSTI2023

Lithium Metal Battery Quality Control via Transformer–CNN Segmentation

Lithium metal battery (LMB) has the potential to be the next-generation battery system because of its high theoretical energy density. However, defects known as dendrites are formed by heterogeneous lithium (Li) plating, which hinders the development and utilization of LMBs. Non-destructive techniques to observe the dendrite morphology often use X-ray computed tomography (XCT) to provide cross-sectional views. To retrieve three-dimensional structures inside a battery, image segmentation becomes essential to quantitatively analyze XCT images. This work proposes a new semantic segmentation approach using a transformer-based neural network called TransforCNN that is capable of segmenting out dendrites from XCT data. In addition, we compare the performance of the proposed TransforCNN with three other algorithms, U-Net, Y-Net, and E-Net, consisting of an ensemble network model for XCT analysis. Our results show the advantages of using TransforCNN when evaluating over-segmentation metrics, such as mean intersection over union (mIoU) and mean Dice similarity coefficient (mDSC), as well as through several qualitatively comparative visualizations.

Quenum, Jerome (ORCID:0000000271265853)↗
DOE OSTI2022

Integrating Deep Learning and Hydrodynamic Modeling to Improve the Great Lakes Forecast

The Laurentian Great Lakes, one of the world’s largest surface freshwater systems, pose a modeling challenge in seasonal forecast and climate projection. While physics-based hydrodynamic modeling is a fundamental approach, improving the forecast accuracy remains critical. In recent years, machine learning (ML) has quickly emerged in geoscience applications, but its application to the Great Lakes hydrodynamic prediction is still in its early stages. This work is the first one to explore a deep learning approach to predicting spatiotemporal distributions of the lake surface temperature (LST) in the Great Lakes. Our study shows that the Long Short-Term Memory (LSTM) neural network, trained with the limited data from hypothetical monitoring networks, can provide consistent and robust performance. The LSTM prediction captured the LST spatiotemporal variabilities across the five Great Lakes well, suggesting an effective and efficient way for monitoring network design in assisting the ML-based forecast. Furthermore, we employed an explainable artificial intelligence (XAI) technique named SHapley Additive exPlanations (SHAP) to uncover how the features impact the LSTM prediction. Our XAI analysis shows air temperature is the most influential feature for predicting LST in the trained LSTM. The relatively large bias in the LSTM prediction during the spring and fall was associated with substantial heterogeneity of air temperature during the two seasons. In contrast, the physics-based hydrodynamic model performed better in spring and fall yet exhibited relatively large biases during the summer stratification period. Finally, we developed a statistical integration of the hydrodynamic modeling and deep learning results based on the Best Linear Unbiased Estimator (BLUE). The integration further enhanced prediction accuracy, suggesting its potential for next-generation Great Lakes forecast systems.

Xue, Pengfei (ORCID:000000025702421X)↗
DOE OSTI2020

High-resolution mapping of spatial heterogeneity in ice wedge polygon geomorphology near Prudhoe Bay, Alaska

It is well known that microtopography associated with ice wedge polygons drives pronounced, meter-scale spatial gradients in hydrologic and ecological processes on the tundra. However, highresolution maps of polygonal geomorphology are rarely available, due to the complexity and subtlety of ice wedge polygon relief at landscape scales. Here we present a sub-meter resolution map of >106 discrete ice wedge polygons across a ~1200 km2 landscape, delineated within a lidar-derived digital elevation model. The delineation procedure relies on a convolutional neural network paired with a set of common image processing operations and permits explicit measurement of relative elevation at the center of each ice wedge polygon. The resulting map visualizes meter- to kilometer-scale spatial gradients in polygonal geomorphology across an extensive landscape with unprecedented detail. This high-resolution inventory of polygonal geomorphology provides rich spatial context for extrapolating observations of environmental processes across the landscape. The map also represents an extensive baseline dataset for quantifying contemporary land surface deformation (i.e., thermokarst) at the survey area, through future topographic surveys.

54 ENVIRONMENTAL SCIENCES↗
NASA NTRS2021

Knowledge Network Embedding of Transcriptomic Data From Spaceflown Mice Uncovers Signs and Symptoms Associated With Terrestrial Diseases

There has long been an interest in understanding how the hazards from spaceflight may trigger or exacerbate human diseases. With the goal of advancing our knowledge on physiological changes during space travel, NASA GeneLab provides an open-source repository of multi-omics data from real and simulated spaceflight studies. Alone, this data enables identification of biological changes during spaceflight, but cannot infer how that may impact an astronaut at the phenotypic level. To bridge this gap, SPOKE, a heterogeneous knowledge graph connecting biological and clinical data from over 30 databases, was used in combination with GeneLab transcriptomic data from six studies. This integration identified critical symptoms and physiological changes incurred during spaceflight.

spaceflight↗
NASA NTRS2017

Monitoring Land Surface Albedo and Vegetation Dynamics Using High Spatial and Temporal Resolution Synthetic Time Series from Landsat and the MODIS BRDF/NBAR/Albedo Product

Seasonal vegetation phenology can significantly alter surface albedo which in turn affects the global energy balance and the albedo warmingcooling feedbacks that impact climate change. To monitor and quantify the surface dynamics of heterogeneous landscapes, high temporal and spatial resolution synthetic time series of albedo and the enhanced vegetation index (EVI) were generated from the 500-meter Moderate Resolution Imaging Spectroradiometer (MODIS) operational Collection V006 daily BRDF (Bidirectional Reflectance Distribution Function) / NBAR (Nadir BRDF-Adjusted Reflectance) / albedo products and 30-meter Landsat 5 albedo and near-nadir reflectance data through the use of the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM). The traditional Landsat Albedo (Shuai et al., 2011) makes use of the MODIS BRDFAlbedo products (MCD43) by assigning appropriate BRDFs from coincident MODIS products to each Landsat image to generate a 30-meter Landsat albedo product for that acquisition date. The available cloud free Landsat 5 albedos (due to clouds, generated every 16 days at best) were used in conjunction with the daily MODIS albedos to determine the appropriate 30-meter albedos for the intervening daily time steps in this study. These enhanced daily 30-meter spatial resolution synthetic time series were then used to track albedo and vegetation phenology dynamics over three Ameriflux tower sites (Harvard Forest in 2007, Santa Rita in 2011 and Walker Branch in 2005). These Ameriflux sites were chosen as they are all quite nearby new towers coming on line for the National Ecological Observatory Network (NEON), and thus represent locations which will be served by spatially paired albedo measures in the near future. The availability of data from the NEON towers will greatly expand the sources of tower albedometer data available for evaluation of satellite products. At these three Ameriflux tower sites the synthetic time series of broadband shortwave albedos were evaluated using the tower albedo measurements with a Root Mean Square Error (RMSE) less than 0.013 and a bias within the range of 0.006. These synthetic time series provide much greater spatial detail than the 500 meter gridded MODIS data, especially over more heterogeneous surfaces, which improves the efforts to characterize and monitor the spatial variation across species and communities. The mean of the difference between maximum and minimum synthetic time series of albedo within the MODIS pixels over a subset of satellite data of Harvard Forest (16 kilometers by 14 kilometers) was as high as 0.2 during the snow-covered period and reduced to around 0.1 during the snow-free period. Similarly, we have used STARFM to also couple MODIS Nadir BRDF-Adjusted Reflectances (NBAR) values with Landsat 5 reflectances to generate daily synthetic times series of NBAR and thus Enhanced Vegetation Index (NBAR-EVI) at a 30-meter resolution. While normally STARFM is used with directional reflectances, the use of the view angle corrected daily MODIS NBAR values will provide more consistent time series. These synthetic times series of EVI are shown to capture seasonal vegetation dynamics with finer spatial and temporal details, especially over heterogeneous land surfaces.

Vegetation Index↗
DOE OSTI2023

Reversibly growing crosslinked polymers with programmable sizes and properties

Growth constitutes a powerful method to post-modulate materials’ structures and functions without compromising their mechanical performance for sustainable use, but the process is irreversible. To address this issue, we here report a growing-degrowing strategy that enables thermosetting materials to either absorb or release components for continuously changing their sizes, shapes, compositions, and a set of properties simultaneously. The strategy is based on the monomer-polymer equilibrium of networks in which supplying or removing small polymerizable components would drive the networks toward expansion or contraction. Using acid-catalyzed equilibration of siloxane as an example, we demonstrate that the size and mechanical properties of the resulting silicone materials can be significantly or finely tuned in both directions of growth and decomposition. The equilibration can be turned off to yield stable products or reactivated again. During the degrowing-growing circle, material structures are selectively varied either uniformly or heterogeneously, by the availability of fillers. Our strategy endows the materials with many appealing capabilities including environment adaptivity, self-healing, and switchability of surface morphologies, shapes, and optical properties. Since monomer-polymer equilibration exists in many polymers, we envision the expansion of the presented strategy to various systems for many applications.

36 MATERIALS SCIENCE↗
DOE OSTI2026

4th TDAMM Workshop White Paper

Time-Domain and Multi-Messenger Astrophysics (TDAMM) is entering a fundamentally new phase characterized by an unprecedented increase in the rate and diversity of astrophysical transient detections. The community is transitioning from a discovery-limited to a follow-up-limited era, driven by major investments across electromagnetic, gravitational-wave, and neutrino observatories. Upcoming facilities such as the Vera C. Rubin Observatory, the Nancy Grace Roman Space Telescope, and wide-field survey instruments will produce a deluge of time-domain alerts, reaching millions of events per night. Simultaneously, upgrades to the gravitational-wave network (LVK O5 and beyond) and neutrino observatories (IceCube Gen2) will significantly increase the detection rates of non-electromagnetic messengers. New high-energy missions and expansions of the InterPlanetary Network (IPN) will further enhance discovery capabilities across the gamma-ray and X-ray regimes. This convergence of capabilities represents a transformative opportunity: for the first time, the community will routinely detect rare and high-impact events across multiple messengers. However, the scientific return from these discoveries will depend critically on the ability to rapidly identify, prioritize, and coordinate follow-up observations across a heterogeneous and globally distributed set of facilities.

79 ASTRONOMY AND ASTROPHYSICS↗
NASA NTRS2000

Dynamic Load-Balancing for Distributed Heterogeneous Computing of Parallel CFD Problems

The developed methodology is aimed at improving the efficiency of executing block-structured algorithms on parallel, distributed, heterogeneous computers. The basic approach of these algorithms is to divide the flow domain into many sub- domains called blocks, and solve the governing equations over these blocks. Dynamic load balancing problem is defined as the efficient distribution of the blocks among the available processors over a period of several hours of computations. In environments with computers of different architecture, operating systems, CPU speed, memory size, load, and network speed, balancing the loads and managing the communication between processors becomes crucial. Load balancing software tools for mutually dependent parallel processes have been created to efficiently utilize an advanced computation environment and algorithms. These tools are dynamic in nature because of the chances in the computer environment during execution time. More recently, these tools were extended to a second operating system: NT. In this paper, the problems associated with this application will be discussed. Also, the developed algorithms were combined with the load sharing capability of LSF to efficiently utilize workstation clusters for parallel computing. Finally, results will be presented on running a NASA based code ADPAC to demonstrate the developed tools for dynamic load balancing.

Ecer, A.↗
DOE OSTI2021

A novel approach to partitioning evapotranspiration into evaporation and transpiration in flooded ecosystems

Reliable partitioning of micrometeorologically measured evapotranspiration (ET) into evaporation (E) and transpiration (T) would greatly enhance our understanding of the water cycle and its response to climate change related shifts in local-to-regional climate conditions and rising global levels of vapor pressure deficit (VPD). While some methods on ET partitioning have been developed, their underlying assumptions make them difficult to apply more generally, especially in sites with large contributions of E. Here, we report a novel ET partitioning method using artificial neural networks (ANNs) in combination with a range of environmental input variables to predict daytime E from nighttime ET measurements. The study uses eddy covariance data from four restored wetlands in the Sacramento-San Joaquin Delta, California, USA, as well as leaf-level T data for validation. The four wetlands vary in their vegetation make-up and structure, representing a range of ET conditions. The ANNs were built with increasing complexity by adding the input variable that resulted in the next highest average value of model testing R 2 across all sites. The order of variable inclusion (and importance) was: VPD > gap-filled sensible heat flux (H_gf) > air temperature (T air ) > friction velocity (u * ) > other variables. The model using VPD, H_gf, T air , and u * showed the best performance during validation with independent data and had a mean testing R 2 value of 0.853 (averaged across all sites, range from 0.728 to 0.910). In comparison to other methods, our ANN method generated T/ET partitioning results which were more consistent with CO 2 exchange data especially for more heterogeneous sites with large E contributions. Further, our method improves the understanding of T/ET partitioning. While it may be particularly suited to flooded ecosystems, it can also improve T/ET partitioning in other systems, increasing our knowledge of the global water cycle and ecosystem functioning.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2025

Knowledge Oriented Graph Unified Transformer (KOGUT) v0.1

KOGUT — Knowledge Oriented Graph Unified Transformer KOGUT implements the Relational Graph Transformer (RelGT) architecture for knowledge graph link prediction in biological domains, with a primary focus on microbial growth media prediction. While the original RelGT (arXiv:2505.10960) targets relational tables, time series, and multi-table databases, KOGUT adapts this architecture for heterogeneous biological knowledge graphs, providing first-in-class AI predictive models for microbial cultivation. Key Adaptations Beyond Original RelGT: - Knowledge Graph Focus: Applied to biological KGs with semantic node types (taxa, chemicals, media, phenotypes, environments) versus generic relational database tables, trained on the KG-Microbe knowledge graph (1.3M entities, 2.9M edges, 24 relation types). - Multimodal Node Encoding: Integrates node labels, categories, descriptions, and synonyms from KG metadata through learned embedding layers—adapting relational column features to graph node attributes with textual semantics. - Extended K-Hop Subgraph Strategy: Optimized neighborhood sampling (3-hop default, configurable up to 200 nodes) tuned for sparse biological networks, building on the original local-global attention framework with biological relation preservation. - Biolink Predicate Preservation: Type-specific transformations for 24 biological edge semantics (occurs_in, consumes, produces, has_phenotype, subclass_of) beyond standard relational foreign keys, enabling multi-relation link prediction. - Inductive Learning Support: Enables zero-shot predictions for novel taxa through feature-based embeddings (temperature, oxygen requirements, gram stain, cell shape), extending the original transductive relational benchmark scope to uncultured microorganisms. CheapSOTA Performance Optimizations (This Distribution): - VQ-EMA Centroid Attention: Vector quantization with exponential moving average for improved global context modeling (+5-10% MRR improvement). - HDF5 Precomputed Data Loading: One-time preprocessing of k-hop subgraphs to eliminate redundant graph traversals (2-5× training speedup). - Distributed Data Parallel Training: Multi-GPU support for scaling to larger knowledge graphs (tested on 4× NVIDIA A100 GPUs at NERSC Perlmutter). - Mixed Precision Training: Automatic mixed precision (AMP) for memory efficiency and faster training. Advantages Over Standard Knowledge Graph Embedding Models: Combines RelGT's proven multi-element tokenization (features, type, hop, structure) with graph-native biological representations, enabling interpretable link prediction across heterogeneous entities that standard embedding models (TransE, RotatE, ComplEx) and table-based transformers cannot directly model. Achieves near-perfect performance on microbial growth media prediction (MRR: 0.9966, Precision@1: 0.9932, Hit@10: 1.0000) while maintaining explainability through attention-based reasoning over biological pathways. Training Data: - KG-Microbe merged knowledge graph: 1,379,337 nodes, 2,960,472 edges - 24 biological relation types including taxonomic hierarchies, metabolic interactions, phenotype associations, and environmental relationships - Primary prediction task: Growth media suitability for microbial taxa (biolink:occurs_in, 50K edges) - Multi-relation capability: Predicts links for any of the 24 relation types, including chemical consumption/production, phenotype associations, and taxonomic classification Citation: Original RelGT Architecture: Dwivedi et al., "Relational Graph Transformer", arXiv:2505.10960, 2025 KOGUT Implementation: Knowledge Oriented Graph Unified Transformer for Microbial Growth Media Prediction Developed at Lawrence Berkeley National Laboratory (LBNL) Trained on NERSC Perlmutter supercomputer

Joachimiak, Marcin [Lawrence Berkeley National Lab↗
DOE OSTI2020

Optimizing long-term monitoring of radiation air-dose rates after the Fukushima Daiichi Nuclear Power Plant

Radiation air dose rates near the Fukushima Daiichi Nuclear Power Plant (FDNPP) have been steadily decreasing over the past eight years since the release of radioactive elements in March 2011. Currently, the radiation monitoring program is expected to transition to long-term monitoring after most of the remediation activities are completed. The main long-term monitoring objectives are to (1) confirm the continuing reduction of contaminant and hazard levels, (2) provide assurance for the public, (3) accumulate the basic datasets for scientific knowledge and future preparation, and (4) detect changes or anomalies in contaminant mobility (if they occur), or any unexpected processes or events. In this work, we have developed a methodology for optimizing the monitoring locations of radiation air dose-rate monitoring. Our approach consists of three steps in order to determine monitoring locations in a systematic manner: (1) prioritizing the critical locations, such as schools or regulatory requirement locations, (2) diversifying locations that cover the key environmental controls that are known to influence contaminant mobility and distributions, and (3) capturing the heterogeneity of radiation air-dose rates across the domain. Therefore, for the second step, we use a Gaussian mixture model to identify the representative locations among multiple environmental variables, such as elevation and land-cover types. For the third step, we use a Gaussian process model to capture and estimate the heterogeneity of air-dose rates across the domain. Employing an integrated dose-rate map derived from Bayesian geostatistical methods as a reference map, we distribute the monitoring locations in such a way as to capture the heterogeneity of the reference map. Our results have shown that this approach allows us to select monitoring locations in a systematic manner such that the heterogeneity of air dose rates is captured by the minimal number of monitoring locations.

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗