Search NASA⌕ Search

SEARCH · Search NASA

Results for “architecture optimization”

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 163 records · Page 9

CRISTAL V2: New package for criticality calculations

A new release of the CRISTAL French criticality calculation package, named CRISTAL V2, has been developed and intensively validated by IRSN and CEA, in collaboration with AREVA. This package was designed to cover the needs of criticality practitioners in the fields of nuclear fuel cycle facilities and transportation of fissile materials. Based on an optimized modular architecture in an updated software development environment, the design of CRISTAL package V2 integrates new generation of cross-section libraries, new calculation schemes consistent with new applications, and up-to-date calculation codes capabilities. Four calculation routes are currently available allowing multi-group and continuous energy calculations. The validation database is composed of more than 3000 critical experiments, in order to cover a wide range of configurations. The CRISTAL V2 package also includes customer services covering a large scope of activities like documentations, hotline, maintenance and training courses. A significant promotion and an increased recognition referred both nationally and internationally are foreseen, including the provision of the CRISTAL V2 package to the OECD / NEA Data Bank. (authors)

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

CRISTAL V2.0: A New-generation criticality package - 14597

A new release of the CRISTAL French criticality calculation package, named CRISTAL V2.0, has been developed and intensively validated by IRSN and CEA, in collaboration with AREVA. This package was designed to cover the needs of criticality practitioners in the fields of nuclear fuel cycle facilities and transportation of fissile materials. Based on an optimized modular architecture in an updated software development environment, the design of CRISTAL package V2.0 integrates new generation of cross section libraries, new calculation schemes consistent with new applications, and up-to-date calculation codes capabilities. Four calculation routes are currently available allowing multi-group and continuous energy calculations. The validation database is composed of more than 3000 critical experiments, in order to cover a wide range of configurations. The CRISTAL V2.0 package also comprises customer services covering a range of areas of activity like documentation, hotline, maintenance and training courses. A wide promotion and an increased recognition referred both nationally and internationally are foreseen, including the provision of the CRISTAL V2.0 package to the OECD / NEA Data Bank. (authors)

42 ENGINEERING↗

Vertex finding in neutrino-nucleus interaction: a model architecture comparison

We compare different neural network architectures for machine learning algorithms designed to identify the neutrino interaction vertex position in the MINERvA detector. The architectures developed and optimized by hand are compared with the architectures developed in an automated way using the package “Multi-node Evolutionary Neural Networks for Deep Learning” (MENNDL), developed at Oak Ridge National Laboratory. While the domain-expert hand-tuned network was the best performer, the differences were negligible and the auto-generated networks performed as well. There is always a trade-off between human, and computer resources for network optimization and this work suggests that automated optimization, assuming resources are available, provides a compelling way to save significant expert time.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

ARENA: Adversary-Resistant Evolving Neural Architectures

Neural networks are becoming the cornerstone for national security prediction tasks. However, designing them requires significant research and trial/error, as they have many hyperparameters, including their computation graph (“architecture”). Neural architecture search (NAS) employs secondary optimizers to search for architectures maximizing objectives like accuracy. Evolutionary algorithms (EAs) are the most used class of optimizer for NAS. However, existing Python libraries for writing EAs limit the complexity of experiments a user can design. In this project, we built ARENA, a Python framework that encodes complex, hyper-realistic EAs. ARENA collects detailed information as it runs and is flexible enough to encode non-EA search algorithms. We tested ARENA on 4 toy optimization problems by encoding 3 search algorithms for each—random search, an EA, and simulated annealing. We also designed an EA that performs NAS on the MNIST dataset. Our experiments suggest the potential for immediate mission impact through solving lab-wide optimization problems.

97 MATHEMATICS AND COMPUTING↗

Invited Paper: Benchmarking and Optimizing Data Movement on Emerging Heterogeneous Architectures

As supercomputers evolve, nodes are continually increasing in complexity. As a result, each generation of parallel systems brings new performance challenges. For instance, on recent systems inter-node communication has outperformed inter-socket, resulting in poor performance of many node-aware communication optimizations. Communication optimizations are critical for the performance and scalability of parallel applications, but are dependent on the parallel architecture, which varies significantly among recent generations of supercomputers. Furthermore, this paper investigates the performance of various paths of data movement on recent generations of systems, and analyzes the increased complexity of communication, particularly on recent heterogeneous systems. The paper also introduces MPI Advance, a communication library that enables optimizations to be created based on benchmark analysis of each emerging system.

benchmarking↗

Capturing in-field root system dynamics with RootTracker

Abstract Optimizing root system architecture offers a promising approach to developing stress tolerant cultivars in the face of climate change, as root systems are critical for water and nutrient uptake as well as mechanical stability. However, breeding for optimal root system architecture has been hindered by the difficulty in measuring root growth in the field. Here, we describe the RootTracker, a technology that employs impedance touch sensors to monitor in-field root growth over time. Configured in a cylindrical, window shutter-like fashion around a planted seed, 264 electrodes are individually charged multiple times over the course of an experiment. Signature changes in the measured capacitance and resistance readings indicate when a root has touched or grown close to an electrode. Using the RootTracker, we have measured root system dynamics of commercial maize (Zea mays) hybrids growing in both typical Midwest field conditions and under different irrigation regimes. We observed rapid responses of root growth to water deficits and found evidence for a “priming response” in which an early water deficit causes more and deeper roots to grow at later time periods. Genotypic variation among hybrid maize lines in their root growth in response to drought indicated a potential to breed for root systems adapted for different environments. Thus, the RootTracker is able to capture changes in root growth over time in response to environmental perturbations.

Aguilar, Jeffrey J.↗

Scaling deep learning workloads: NVIDIA DGX-1/Pascal and Intel Knights Landing

Deep Learning (DL) algorithms have become ubiquitous in data analytics. As a result, major computing vendors --- including NVIDIA, Intel, AMD, and IBM --- have architectural road-maps influenced by DL workloads. Furthermore, several vendors have recently advertised new computing products as accelerating large DL workloads. Unfortunately, it is difficult for data scientists to quantify the potential of these different products. This paper provides a performance and power analysis of important DL workloads on two major parallel architectures: NVIDIA DGX-1 (eight Pascal P100 GPUs interconnected with NVLink) and Intel Knights Landing (KNL) CPUs interconnected with Intel Omni-Path or Cray Aries. Our evaluation consists of a cross section of convolutional neural net workloads: CifarNet, AlexNet, GoogLeNet, and ResNet50 topologies using the Cifar10 and ImageNet datasets. The workloads are vendor-optimized for each architecture. Our analysis indicates that although GPUs provide the highest overall performance, the gap can close for some convolutional networks; and the KNL can be competitive in performance/watt. We find that NVLink facilitates scaling efficiency on GPUs. However, its importance is heavily dependent on neural network architecture. Furthermore, for weak-scaling --- sometimes encouraged by restricted GPU memory --- NVLink is less important.

Gawande, Nitin A.↗

Simulation-Based Parameter Optimization Framework for Large-Scale Hybrid Smart Grid Communications Systems Design

The design of reliable, dynamic, fault-tolerant hybrid smart grid communication networks is a challenge to achieve for autonomous power grids. Hybrid networks use different communications technologies for different area networks. A simulation-based parameter optimization framework is proposed to tune parameters of hybrid communication technologies to achieve the optimal network performance. It consists of three main components: a parallel executor used to speedup a list of simulations; a sampler running simulations using the parallel executor at each generation; and a hybrid stochastic optimization algorithm for tuning configurable parameters of hybrid designs and applications. The proposed hybrid metaheuristic optimization algorithm combines an evolutionary algorithm with a gradient method to quickly achieve an approximately global optimum solution. Three optimization test functions are employed to train the adjustable parameters of the hybrid algorithm. Results show the proposed parameter optimization framework can help the designer choose the right hybrid architecture with an optimal parameter set for a large-scale broadband PLC-WiMAX hybrid smart grid communication network.

24 POWER TRANSMISSION AND DISTRIBUTION↗

HARMONY: Large-Scale Architecture Search for Efficient Hybrid Language Models

As large language models scale to trillions of parameters, their computational and memory requirements present critical challenges for efficient training and deployment. While Mixture of Experts (MoE) architectures enable efficient scaling through sparse parameter activation, and state-space models like Mamba offer linear-time complexity, principled methods for combining these paradigms remain undeveloped. We introduce HARMONY (Hybrid Architecture Research for Mamba, Optimized with Neural efficiencY), a multi-objective evolutionary neural architecture search framework for discovering efficient hybrid language models that integrate Transformer attention mechanisms, Mixture-of-Experts routing, and Mamba state-space components. Through large-scale distributed search using 16,384 MI250X GPUs on the Frontier supercomputer, HARMONY explores a comprehensive design space encompassing six attention variants (MHA, MQA, GQA, MLA, SWA, and Mamba-2), variable MoE configurations with both routed and shared experts, and extensive Mamba hyperparameters. Our framework discovers heterogeneous architectures that balance training performance with computational efficiency through multi-objective optimization incorporating latency penalties and fitness-based selection. Analysis of discovered architectures reveals that optimal hybrid designs favor heterogeneous component mixing rather than homogeneous patterns, with Mamba-2 and Multi-Head Latent Attention (MLA) emerging as preferred mechanisms. Discovered architectures demonstrate superior training efficiency: our best configuration achieves a final perplexity of 1.0874 with 2.38B parameters while processing 4,320 tokens/second, outperforming significantly larger manually designed models. Full-scale evaluation shows HARMONY's top architectures achieve better loss trajectories than equivalently-sized models using state-of-the-art configurations including Mixtral, Jamba, and Samba. Additionally, we demonstrate 91% weak scaling efficiency when training discovered 36B-parameter models across 1,024 GPUs. HARMONY is released as an open framework with comprehensive tools for building and training hybrid models using expert-data-pipeline parallelism, democratizing access to automated architecture design for next-generation language models.

Herron, Emily [ORNL] (ORCID:0000000273008172)↗

High Energy Yield Distributed Architecture for Large Commercial and Utility-Scale PV Systems

Photovoltaic energy has come a long way from its early days when small cells were fabricated in the labs in the early 50s. While it took a long time for PV modules to mature into what we now see in the market and in rooftops across the world, the power electronics has been lagging behind & was always the weakest point in the PV plant, being responsible for the majority of failures & robbing the PV plants of their full potential by losing precious Watts in the process of converting power from DC to AC. Since the early days of PV power, it has been recognized that the modules need to operate at their maximum power point in order to extract as much power as is available from the PV modules at a given time of day and year. Numerous maximum power point devices were developed to do just this. While effective at extracting as much power as they can, soon after there was a realization that in a PV installation where a central inverter is used, many losses occur, hence energy yield losses. This has led to thinking about alternative architectures where the maximum power point tracking (MPPT) is performed much closer to the modules rather than at a central inverter. Several architectures were proposed ranging from a microinverter, microconverter, string inverter, string optimizer, and to a multi-string inverter or optimizer. While these architectures promise to increase the energy yield by 5% to 25% (depending where MPPT is located), & recovering losses due to shading, soiling, mismatch … there were no real studies and implementations in the field to corroborate these numbers. The program that GE Global Research proposed & received funding for is aimed at a type of distributed architectures, namely the dc architectures where only the MPPT function is distributed across the PV plant. Further, the project is concerned with choosing the right architecture for large scale commercial & utility scale plants but not residential installations. Findings from Task 1- Define Requirements & Develop Architecture- showed that for large scale commercial & utility PV plants, locating the MPPT at the string level with a string size of >3kW is the most cost effective solution & that the increase in energy yield of 6% to 8% is more than enough to make up for the increased cost due to additional power electronics. Our study also shows that in addition to an increase in energy yield, having a dc/dc converter at the string level brings additional benefits such as increased array visibility, allowing for increased monitoring and diagnostics; high and constant dc bus voltage which allows for reducing losses further and optimizing the inverter; and ability to detect ground and arc faults. Task 2 focused on designing a suitable dc/dc converter that meets the requirements defined in Task 1: Efficiency, Cost, & Reliability. For this task, we surveyed and simulated all candidate topologies known by the practitioners in the field. We also generated a novel topology, simulated it, & compared it to existing topologies. Our novel topology being a partial power processing converter came out on top due to its simplicity (fewer parts), it high efficiency, its reliability, and its lower cost. In the remaining tasks of the project, we built the hardware, tested it first in the lab, then in the field and once we worked out all the bugs and issues and were confident our system was robust, we installed it on the rooftop and started collecting the data by mid-Q3 2012. We shared our work with the scientific community in the form of conference & Journal papers, hence contributing to the general understanding of this technology area. The methods, techniques, and hardware developed under this project have all been tested in the field over long periods of time, and under many conditions and have been shown to be robust and reliable. The converters built and developed under this program have been in operation in the field since mid-June 2012, and no hardware failures have been recorded. While some software issues have been seen, none have been showstoppers. The cost analysis of the hardware showed that it meets the cost target set in the requirement document developed in the trade-off study. Improving the performance of solar installations for commercial and utility scale plants contributes to reducing their cost of ownership, hence spreading them to a larger section of the society and making them affordable. Our project contributed to the understanding of the distributed architectures, showed that PV plants’ cost of electricity can be reduced through additional power electronics which allows for harvesting more energy from the plant. When the project started in 2009, the cost of PV energy was still high, hence there was a real need to increase the energy yield, hence reduce the levelized cost of electricity (LCOE). While costs have come down significantly, the need for increasing the energy yield is still relevant, hence the project results, hardware, and techniques developed are still pertinent today. By taking this project from concept to rooftop demonstration, we showed that the technology is mature enough, robust enough, and has the potential to be used by many more customers, hence benefiting the public in general through the spread of PV energy, which ultimately contributes to reducing our dependency on fossil fuel, improving the environment, and creating more jobs.

14 SOLAR ENERGY↗

Understanding implications of cathode architecture on energy density of solid-state batteries

Next generation solid-state batteries (SSB) will need to leverage high voltage cathodes, as well as metallic anodes to achieve the realistic performance targets necessary to replace liquid electrolyte-based batteries in cutting-edge applications including electric vehicles. However, limitations arising from mass and charge transports, kinetics and chemo-mechanical degradation at the electrode | electrolyte interface limit the performance of present day SSBs. Optimizing composite cathode architecture, which is an integral part of solid-state batteries, is vital to realize the high-energy density and high-performance goals for next-generation solid-state batteries. Additionally, cathode architecture needs to be optimized for high loadings of active material, well-percolated ion and electron transport pathways and increased resilience against electrochemical stresses. This paper provides a first report of framework for geometric modeling of composite cathode architectures and evaluates the impact of cathode architecture on cell-level energy density using hierarchical models. Packing around primary and secondary active material particles are simulated for a range of active material particle size and solid electrolyte size distributions in the composite cathode. Impact of packing architecture on processing parameters of a given cathode composition and thickness, as well as on achievable energy density is evaluated for a range of commonly used solid electrolyte and cathode materials. Overall, the proposed framework offers a facile exploratory methodology for establishing initial metrics for scalable processing of practical and competent SSBs.

25 ENERGY STORAGE↗

Surrogate Model Based Optimization for Finding Robust Deep Learning Model Architectures

Deep Learning (DL) models are increasingly used throughout the sciences. However, their performance and usefulness depend greatly on their architecture which is defined by hyperparameters such as the number of nodes, layers, the learning rate, etc. Tuning these hyperparameters is time-consuming because evaluating their performance requires a lengthy training step. Stochastic optimizers used in training lead to performance variability and potentially prediction reliability issues. In this talk, we will describe an automated optimization method based on surrogate models and active learning strategies for tuning DL model architectures. We take into account the prediction variability with the goal to identify architectures that make reliable and robust predictions. We demonstrate our developments on an application arising in particle physics.

deep learning↗

A Nonlinear Direct Power Controller for a Three-Phase Grid-Connected Inverter with Online Parameter Update for PV Application

This work proposes a nonlinear direct power controller for a three-phase grid-connected inverter for photovoltaic application. Grid-connected converters for solar applications require advanced control architectures to ensure optimal operation and extraction of maximum power from the photovoltaic arrays. Therefore, to ensure this, a nonlinear power controller architecture has been proposed to extract the maximum power from the solar arrays and to achieve the most optimal operation. To further improve the performance, a parameter update scheme has been utilized which is based on model reference adaptive system. The latter helps to accomplish not only good dynamic response but also excellent tracking performance. The overall control come estimation architecture is verified based on computer simulations via MATLAB/Simulink and PLECS domain and various important case study results are presented.

direct power control↗

Differentiable Neural Architecture, Mixed Precision and Accelerator Co-Search

Quantization, effective Neural Network architecture, and efficient accelerator hardware are three important design paradigms to maximize accuracy and efficiency. Mixed Precision Quantization is a process of assigning different precision to different Neural Network layers for optimized inference. Neural Architecture Search (NAS) is a process of automatically designing the neural network for a task and can also be extended to search for the precision of each weight and activation matrix. In this paper, we develop the following three methods: (i) Fast Differentiable Hardware-aware Mixed Precision Quantization Search method to find optimal precision, (ii) Joint Differentiable hardware-aware Architecture and Mixed Precision Quantization Co-search, (iii) Joint Accelerator, Architecture, and Precision triple co-search to find best possibilities in all the three worlds. We demonstrate the effectiveness of our proposed methods targeting Bitfusion accelerator by searching mixed precision models on MobilenetV2. We achieve better accuracy-latency trade-off models than the manually designed and previously proposed search methods.

97 MATHEMATICS AND COMPUTING↗