Search NASA⌕ Search

SEARCH · Search NASA

Results for “Building Dataset”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,063 records · Page 59

AI Model Benchmarking for Nonproliferation Applications: Steel Thread Benchmarking Task Force Technical Report (Rev. 2)

Steel Thread is a NA-22 venture that seeks to build trustworthy, reliable AI models that can be used in a wide variety of nonproliferation tasks. A key aspect of building these models is developing appropriate benchmarks and evaluation methods, which will enable the venture to identify and adapt models to provide the most value in the nonproliferation domain. Benchmarks must be relevant to key tasks in this domain, such as question answering, information retrieval, document summarization and classification, consensus analysis, and image and data analysis. This report 1) provides an overview of benchmark design, evaluation, and challenges; 2) reviews a variety of open benchmarks, with a focus on language models and tasks; and 3) identifies benchmarks that are most relevant to Steel Thread. This report is intended to serve as a basis for further efforts to classify and evaluate benchmarks and their correlation with success on nonproliferation-specific tasks. The Steel Thread venture has defined benchmarks to be a particular combination of a dataset (or datasets) and a metric (or metrics) conceptualized as representing one or more specific tasks or sets of abilities for a specific modality. It is adopted by a research community as a shared framework for comparing methods.1 It includes 1) Data: Labeled (a designated subset not used for training, which could be all the data), 2) Metric: A way to quantify performance, 3) Task/Ability: What the benchmark is testing, 4) Protocol: A structured and repeatable evaluation process, 5) Baseline/Reference Model: For comparison; could be statistical, rule-based, SME-derived, or another model, and 6) Maintenance Plan: to update with new information over time; important for long-term utility. For further clarity, the definition includes what a benchmark, in this context, is not. It is not a corpus of training data, specific to a model (it is intended to apply to a range of models), a universal evaluation of performance, a guarantee that the ‘top’ model on the leaderboard will be the best fit for every specific use case, an all-encompassing proof of a model’s universal quality, nor is it a one-size-fits-all measure of success. It does not cover every real-world constraint (like operational, ethical, or cost considerations), a systems integration test, or a unit test. This definition was inspired by and resulted from discussions within the Steel Thread Benchmarking Task Force. This group was formed to define what we would mean as a benchmark within Steel Thread but persisted as the need to develop a thorough understanding of the large and expanding existing benchmarking space. This technical report is a result of the group’s divide and conquer approach to exploring this space. The release of benchmarks might not be progressing as quickly as model development, but it is moving very fast, as many benchmarks quickly become saturated, when state-of-the-art models score so close to the benchmark’s ceiling that their results are virtually indistinguishable. At that point, the test no longer differentiates between new systems, so researchers usually stop reporting scores as the benchmark no longer informs about improvements from the next generation of models. In the OpenAI announcement of GPT-5, they reported results on six flagship public benchmarks (AIME 2025, SWE-bench Verified, Aider Polyglot, MMMU, HealthBench Hard, GPQA) but the full system-card covers roughly thirty-five separate evaluations, comprising hundreds of test task items in total. There have been some efforts to summarize benchmarks in specific fields, like for text-to-image generation, but these surveys have had a narrow methodology scope. Therefore, a comprehensive survey of all benchmarks or even all benchmarks that could be relevant to Steel Thread is outside of the scope of this report. We chose some specific benchmarks to investigate in detail.

97 MATHEMATICS AND COMPUTING↗

Resolving the Discrepancies Between Empirical and Rayleigh Charge Limiting Models for Globular Proteins

Starting with the Rayleigh charge limiting model, a slightly different approach is used to account for the well-known discrepancy that exists between the said model and experimental ESI MS data for globular proteins. It is shown using published datasets that for globular proteins, the mass density ρ exhibits a weak second-order dependence on its mass M, according to ρ(M)∝M{sup -α}, α ~ 0.14. A direct equivalence established between ESI MS and x-ray techniques suggests a minimum but critical surface tension of 15.6 ± 5.2 mN/m for the droplet at the liquid-to-gas phase transition point. The packing density factor η for globular proteins is believed to lie between 1 (very tightly packed) and 4.6 (less tight, natively packed). While the Rayleigh charge limiting model has been linked historically to the CRM (J. Chem. Phys. 49:2240–2249, 1968; Anal. Chim. Acta 406:93–104, 2000), this paper does not expressly seek to justify the CRM, but rather uses empirical data and existing knowledge across subfields to help build a consistent picture of ESI MS phenomena that might be difficult to explain otherwise. These results would be useful in molecular dynamics (MD) simulations, understanding liquid-to-gas phase transitions and in opening up new routes for cross-calibration between ESI MS, IM MS, NMR and x-ray crystallography studies. .

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

The Net Carbon Flux due to Deforestation and Forest Re-Growth in the Brazilian Amazon: Analysis using a Process-Based Model

We developed a process-based model of forest growth, carbon cycling, and land cover dynamics named CARLUC (for CARbon and Land Use Change) to estimate the size of terrestrial carbon pools in terra firme (non-flooded) forests across the Brazilian Legal Amazon and the net flux of carbon resulting from forest disturbance and forest recovery from disturbance. Our goal in building the model was to construct a relatively simple ecosystem model that would respond to soil and climatic heterogeneity that allows us to study of the impact of Amazonian deforestation, selective logging, and accidental fire on the global carbon cycle. This paper focuses on the net flux caused by deforestation and forest re-growth over the period from 1970-1998. We calculate that the net flux to the atmosphere during this period reached a maximum of approx. 0.35 PgC/yr (1PgC = 1 x 10(exp I5) gC) in 1990, with a cumulative release of approx. 7 PgC from 1970- 1998. The net flux is higher than predicted by an earlier study by a total of 1 PgC over the period 1989-1 998 mainly because CARLUC predicts relatively high mature forest carbon storage compared to the datasets used in the earlier study. Incorporating the dynamics of litter and soil carbon pools into the model increases the cumulative net flux by approx. 1 PgC from 1970-1998, while different assumptions about land cover dynamics only caused small changes. The uncertainty of the net flux, calculated with a Monte-Carlo approach, is roughly 35% of the mean value (1 SD).

Hirsch, A. I.↗

Information Entropy as Quantifier of Potential Predictability in the Tropical Indo-Pacific Basin

Global warming is posed to modify the modes of variability that control much of the climate predictability at seasonal to interannual scales. The quantification of changes in climate predictability over any given amount of time, however, remains challenging. Here we build upon recent advances in non-linear dynamical systems theory and introduce the climate community to an information entropy quantifier based on recurrence. The entropy, or complexity of a system is associated with microstates that recur over time in the time-series that define the system, and therefore to its predictability potential. A computationally fast method to evaluate the entropy is applied to the investigation of the information entropy of sea surface temperature in the tropical Pacific and Indian Oceans, focusing on boreal fall. In this season the predictability of the basins is controlled by two regularly varying non-linear oscillations, the El Niño-Southern Oscillation and the Indian Ocean Dipole. We compute and compare the entropy in simulations from the CMIP5 catalog from the historical period and RCP8.5 scenario, and in reanalysis datasets. Discrepancies are found between the models and the reanalysis, and no robust changes in predictability can be identified in future projections. The Indian Ocean and the equatorial Pacific emerge as troublesome areas where the modeled entropy differs the most from that of the reanalysis in many models. A brief investigation of the source of the bias points to a poor representation of the ocean mean state and basins' connectivity at the Indonesian Throughflow.

54 ENVIRONMENTAL SCIENCES↗

NBIC Biofeeds: A Digital Tool for Open Source Biosurveillance across Federal Agencies

The National Biosurveillance Integration Center (NBIC) is developing a scalable, flexible open source data collection, analysis, and dissemination tool to support biosurveillance operations by the U.S. Department of Homeland Security (DHS) and its federal interagency partners. The NBIC integrates, analyzes, and distributes key information about health and disease events to help ensure the nation’s responses are well-informed, save lives, and minimize economic impact. NBIC serves as a bridge between Federal, State, Local, Territorial, and Tribal entities to conduct biosurveillance across human, animal, plant, and environmental domains. The integration of information enables early warning and shared situational awareness of biological events to inform critical decisions directing response and recovery efforts. To meet its mission objectives, NBIC utilizes a variety of datasets, including open source information, to provide comprehensive coverage of biological events occurring across the globe. NBIC Biofeeds is a digital tool designed to improve the efficiency of reviewing and analyzing large volumes of open source reporting by biosurveillance analysts on a daily basis; moreover, the system provides a mechanism to disseminate tailored feeds allowing NBIC to better meet the specific information needs of individual, interagency partners. The tool is currently under development by the Department of Energy (DOE), Pacific Northwest National Laboratory (PNNL)and it is in a testing and evaluation phase supported by NBIC biosurveillance subject matter experts. Integration with the Defense Threat Reduction Agency (DTRA), Biosurveillance Ecosystem(BSVE) is also underway. NBIC Biofeeds Version 1 is expected to be fully operational in Fiscal Year 2017. The PNNL is applying agile methodology to streamline the build of NBIC Biofeeds to specifications required for operational use by NBIC and its federal interagency partners. Biosurveillance, analytics, and system engineering subject matter experts provide guidance on the implementation of features in the tool to ensure functionality aligns with operational workflows and production support. PNNL is leveraging software from a previous government effort to repurpose the technology to meet NBIC needs. NBIC Biofeeds incorporates the open source, document-orientated MongoDB database to capture user- and system-generated metadata on hundreds of thousands of records, in part, to establish baselines to aid prospective and retrospective analysis on emerging biological events. NBIC Biofeeds integrates a biosurveillance taxonomy (uniquely developed by NBIC),which includes input from interagency partners to recognize critical characteristics of a biological event. In NBIC Biofeeds Version1, metadata capture of reported events is done manually by NBIC analysts; however, moving forward in Version 2, the tool will be further automated to flag significant reporting on biological events with a human remaining in the loop to confirm the validity of the system-generated tags. To serve as a one-stop tool for open source biosurveillance, NBIC Biofeeds automatically harvests information from thousands of websites, utilizing third party aggregators, paid subscriptions to data feeds, and scraping of high priority sources. Users can develop desired queries for automatic updating, leverage a unique review and curation mechanism, and further analyze data from topical, geographic, and temporal visualization features in the tool. To meet NBIC’s information sharing needs, the tool allows for design of tailored RSS feeds and electronic message-based delivery of analysis on biological events, intended for recipients in the government with unique missions around human, animal, plant, and environmental health. Through current testing and evaluation – underway by biosurveillance subject matter experts – NBIC Biofeeds is demonstrating value in supporting open source biosurveillance by the Center for more rapid recognition and sharing of key event characteristics. Centralizing access and analysis of this data set into a single system is increasing the efficiency of daily, global biosurveillance, while enhancing the value of information identified through use of the querying, curation, and production support features in the tool.

59 BASIC BIOLOGICAL SCIENCES↗

The SONATA data format for efficient description of large-scale network models

Increasing availability of comprehensive experimental datasets and of high-performance computing resources are driving rapid growth in scale, complexity, and biological realism of computational models in neuroscience. To support construction and simulation, as well as sharing of such large-scale models, a broadly applicable, flexible, and high-performance data format is necessary. To address this need, we have developed the Scalable Open Network Architecture TemplAte (SONATA) data format. It is designed for memory and computational efficiency and works across multiple platforms. The format represents neuronal circuits and simulation inputs and outputs via standardized files and provides much flexibility for adding new conventions or extensions. SONATA is used in multiple modeling and visualization tools, and we also provide reference Application Programming Interfaces and model examples to catalyze further adoption. SONATA format is free and open for the community to use and build upon with the goal of enabling efficient model building, sharing, and reproducibility.

59 BASIC BIOLOGICAL SCIENCES↗

Revisiting matrix-based inversion of scanning mobility particle sizer (SMPS) and humidified tandem differential mobility analyzer (HTDMA) data

Abstract. Tikhonov regularization is a tool for reducing noise amplification during data inversion. This work introduces RegularizationTools.jl, a general-purpose software package for applying Tikhonov regularization to data. The package implements well-established numerical algorithms and is suitable for systems of up to ∼ 1000 equations. Included is an abstraction to systematically categorize specific inversion configurations and their associated hyperparameters. A generic interface translates arbitrary linear forward models defined by a computer function into the corresponding design matrix. This obviates the need to explicitly write out and discretize the Fredholm integral equation, thus facilitating fast prototyping of new regularization schemes associated with measurement techniques. Example applications include the inversion involving data from scanning mobility particle sizers (SMPSs) and humidified tandem differential mobility analyzers (HTDMAs). Inversion of SMPS size distributions reported in this work builds upon the freely available software DifferentialMobilityAnalyzers.jl. The speed of inversion is improved by a factor of ∼ 200, now requiring between 2 and 5 ms per SMPS scan when using 120 size bins. Previously reported occasional failure to converge to a valid solution is reduced by switching from the L-curve method to generalized cross-validation as the metric to search for the optimal regularization parameter. Higher-order inversions resulting in smooth, denoised reconstructions of size distributions are now included in DifferentialMobilityAnalyzers.jl. This work also demonstrates that an SMPS-style matrix-based inversion can be applied to find the growth factor frequency distribution from raw HTDMA data while also accounting for multiply charged particles. The outcome of the aerosol-related inversion methods is showcased by inverting multi-week SMPS and HTDMA datasets from ground-based observations, including SMPS data obtained at Bodega Marine Laboratory during the CalWater 2/ACAPEX campaign and co-located SMPS and HTDMA data collected at the US Department of Energy observatory located at the Southern Great Plains site in Oklahoma, USA. Results show that the proposed approaches are suitable for unsupervised, nonparametric inversion of large-scale datasets as well as inversion in real time during data acquisition on low-cost reduced-instruction-set architectures used in single-board computers. The included software implementation of Tikhonov regularization is freely available, general, and domain-independent and thus can be applied to many other inverse problems arising in atmospheric measurement techniques and beyond.

54 ENVIRONMENTAL SCIENCES↗

Standardizing Interfaces for External Access to Data and Processing for the NASA Ozone Product Evaluation and Test Element (PEATE)

NASA's traditional science data processing systems have focused on specific missions, and providing data access, processing and services to the funded science teams of those specific missions. Recently NASA has been modifying this stance, changing the focus from Missions to Measurements. Where a specific Mission has a discrete beginning and end, the Measurement considers long term data continuity across multiple missions. Total Column Ozone, a critical measurement of atmospheric composition, has been monitored for'decades on a series of Total Ozone Mapping Spectrometer (TOMS) instruments. Some important European missions also monitor ozone, including the Global Ozone Monitoring Experiment (GOME) and SCIAMACHY. With the U.S.IEuropean cooperative launch of the Dutch Ozone Monitoring Instrument (OMI) on NASA Aura satellite, and the GOME-2 instrumental on MetOp, the ozone monitoring record has been further extended. In conjunction with the U.S. Department of Defense (DoD) and the National Oceanic and Atmospheric Administration (NOAA), NASA is now preparing to evaluate data and algorithms for the next generation Ozone Mapping and Profiler Suite (OMPS) which will launch on the National Polar-orbiting Operational Environmental Satellite System (NPOESS) Preparatory Project (NPP) in 2010. NASA is constructing the Science Data Segment (SDS) which is comprised of several elements to evaluate the various NPP data products and algorithms. The NPP SDS Ozone Product Evaluation and Test Element (PEATE) will build on the heritage of the TOMS and OM1 mission based processing systems. The overall measurement based system that will encompass these efforts is the Atmospheric Composition Processing System (ACPS). We have extended the system to include access to publically available data sets from other instruments where feasible, including non-NASA missions as appropriate. The heritage system was largely monolithic providing a very controlled processing flow from data.ingest of satellite data to the ultimate archive of specific operational data products. The ACPS allows more open access with standard protocols including HTTP, SOAPIXML, RSS and various REST incarnations. External entities can be granted access to various modules within the system, including an extended data archive, metadata searching, production planning and processing. Data access is provided with very fine grained access control. It is possible to easily designate certain datasets as being available to the public, or restricted to groups of researchers, or limited strictly to the originator. This can be used, for example, to release one's best validated data to the public, but restrict the "new version" of data processed with a new, unproven algorithm until it is ready. Similarly, the system can provide access to algorithms, both as modifiable source code (where possible) and fully integrated executable Algorithm Plugin Packages (APPs). This enables researchers to download publically released versions of the processing algorithms and easily reproduce the processing remotely, while interacting with the ACPS. The algorithms can be modified allowing better experimentation and rapid improvement. The modified algorithms can be easily integrated back into the production system for large scale bulk processing to evaluate improvements. The system includes complete provenance tracking of algorithms, data and the entire processing environment. The origin of any data or algorithms is recorded and the entire history of the processing chains are stored such that a researcher can understand the entire data flow. Provenance is captured in a form suitable for the system to guarantee scientific reproducability of any data product it distributes even in cases where the physical data products themselves have been deleted due to space constraints. We are currently working on Semantic Web ontologies for representing the various provenance information. A new web site focusing on consolidating informaon about the measurement, processing system, and data access has been established to encourage interaction with the overall scientific community. We will describe the system, its data processing capabilities, and the methods the community can use to interact with the standard interfaces of the system.

Tilmes, Curt A.↗

Onboard Hyperspectral Image Classification via Transfer Learning for Communication-Limited Spacecraft

Employing deep-learning and artificial-intelligence (AI) techniques onboard spacecraft can dramatically improve priority data selection to ensure more effective use of the available downlink. However, deployment of effective deep-learning models requires significant training on the ground, which may not be feasible, due to limited data available in an unexplored environment. Therefore, this research explores building robust classification models for onboard data processing where training data is highly limited using transfer-learning techniques. In this paper, we focus on the use case of hyperspectral imaging for remote sensing, a domain where the high dimensionality of the data from the sensor can rapidly saturate the downlink bandwidth. With this bottleneck, there is an impending need to autonomously and robustly classify data onboard to optimize downlink of high-impact measurements, thus maximizing the scientific utility per bit transmitted to the ground. This paper examines the use of deep neural networks onboard for hyperspectral image classification in a communication-limited scenario to analyze how the models perform with limited training data. The use of transfer learning can ameliorate the issue of poor generalization by transferring features learned from training on a large source dataset for one classification task to the target classification task with limited training data. For two deep-learning models from literature, we compare the accuracy of the models trained using transfer learning to models trained from scratch using a random weight initialization with varying amounts of training data. We demonstrate the feasibility and performance of running inference of the deep-learning models on representative flight-like hardware.

Advanced Avionics, Machine Learning, Data Processi↗

The GeoGLOWS Project: Essential Water Variables and Observations – Expected Value Chain, Products, Examples

Essential Water Variables (EWV) were developed over the last decade to ensure that key datasets were made available for closing the global water budget, and addressing the many international actions that require water data. Satellite, surface, and numerical model data all have a role in providing EWVs, and it is a key point that these data must be merged and then fed into decision support applications to answer all the needs of the panoply of user communities. The more specialized the applications become, the higher the demand for development and maintenance resources. Some examples of applications include the NASA Goddard Giovanni, which provides on-line data access and simple analysis; the Global Flood Monitor, which focuses on precipitation-driven flooding; the NASA Landslide Hazard Assessment for Situational Awareness, which similarly treats landslides; and the Global Streamflow Forecasting Project, which provides multiple decision support interfaces, as well as a general toolkit for building new interfaces.

Essential Water Variables↗

Dosimetric Predictors of Radiation-Induced Vaginal Stenosis After Pelvic Radiation Therapy for Rectal and Anal Cancer

Purpose: Although vaginal stenosis (VS) is a recognized toxicity in women who receive pelvic radiation therapy (RT), the relationship between RT dose and the volume and extent of toxicity has not been analyzed. We modeled this relationship to identify predictors of VS. Methods and Materials: We evaluated 54 women, aged 29 to 78 years, who underwent pelvic RT for rectal or anal cancer during 2008 to 2011 and were enrolled in a prospective study evaluating vaginal dilator use. Maximum dilator size was measured before RT (baseline) and 1 month and 12 months after RT. Dilator use was initiated at 1 month. The difference (D) in dilator size before and after RT was recorded. Those with D ≤−1 were classified as having VS (n=35); those with D ≥0 were classified as having no VS (n=19 at 1 month). Dose-volume parameters were extracted, and the generalized equivalent uniform dose (gEUD) was used to build a predictive model. Results: The mean vaginal doses were 50.0 Gy and 36.8 Gy for anal and rectal cancer patients, respectively. One month after RT, a gEUD model using a wide range of a values suggests that sparing of vaginal volume to a low dose may be important. When gEUD (a = −1) was <35 Gy and the mean vaginal dose was <43 Gy, severe VS was reduced (P=.02). A 1-year analysis suggests increasingly negative D values with increasing mean dose. However, patients with compliance <40% were more likely to have toxicity. Conclusions: Vaginal stenosis is influenced by multiple RT dose-volume characteristics. Mean dose and gEUD constraints together may reduce the risk of severe VS. Patients receiving higher mean vaginal doses should have greater compliance with dilator therapy to minimize risk of toxicity. Further validation with independent datasets is needed.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Plasma pressure profiles in a sheared-flow-stabilized Z-pinch

We report the plasma pressure reached inside the central plasma column of a sheared-flow-stabilized Z-pinch using Thomson scattering measurements. Building on previously reported experimental results and the analysis methods established for the high temperature and moderate density plasmas generated on the FuZE device, we show evidence of a central plasma region with higher electron temperature and density, which is consistent with a pinch behavior. Elevated electron temperatures up to 2.25 ± 0.8 keV and densities up to (4.9±0.2)×1017 cm−3 are observed to temporally coincide with the fusion neutron production from the plasma. Reconstructed plasma pressure profiles highlight the presence of a several millimeter-wide column with elevated pressure whose location varies shot-to-shot. The plasma pressure rises as neutron production from the deuterium plasma increases, reaching a peak value of 2.6 kBar. This peak value is consistent with a radially force-balanced pinch equilibrium model based on the measured ∼320 kA pinch current. Complete datasets were obtained at two axial locations, 10 and 20 cm axial position from the tip of the central electrode, which corroborate the estimated neutron source axial lengths.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

“Shoulda, Coulda, Woulda”: Conceptualizing the Differences in Trust Between Human-Human Teaming and Human-Machine Teaming

Intelligent decision support systems (IDSSs) are machine teammates designed to facilitate better human decision-making in high-consequence domains such as health care, power grid operations, and fraud detection. IDSSs identify patterns in datasets and provide intelligent decision-making recommendations to human teammates. However, previous research indicates that humans often trust IDSS recommendations less than the recommendations from their human teammates, even when the machine teammate is more accurate. To conceptualize why trust differs, we review the literature surrounding trust, error, and predictability. Then, we compile and compare participant trust ratings and decision-making in an abridged systematic review of previous studies manipulating teammate type, error rate, and error type. Finally, we conduct a content analysis of participants’ qualitative responses to trust queries from a survey on generative language models. Results suggest that humans may trust IDSS teammates less than other human teammates because of differences in (1) interaction complexity, (2) blame attribution, and (3) swift trust. We conclude that human factors practitioners should collaborate with data scientists and domain experts to build and maintain trust in IDSSs by anthropomorphizing algorithms, matching mental models, and considering individual differences.

97 MATHEMATICS AND COMPUTING↗

Charting the unknown: A dive into the world of standards mapping

When you hear standards mapping what do you think about? For the Secure Software Central (SSC) project this means comparing our process to the National Institute of Standards and Technology’s (NIST) security control catalog. So how is it done? The project completed this summer can be broken down into three simple phases. Phase one consists of mapping the SSC process to the NIST’s security control catalog. While phase two entails mapping mitigations from a threat profile to NIST. Lastly phase three requires the creation of a knowledge base allowing for the reuse of controls across numerous projects This project presented a few different challenges: locating the correct control within the NIST catalog while correctly matching the proposed mitigations from the SSC team, accurately leveling the mitigation to the security level of the system, creating a toolbox to showcase a dataset of mitigations and standards to be used in current and future projects. each challenge allowed for opportunities in growth and understanding of how security controls can map to a governing set of standards. These standards maps are a crucial element to support the insights provided by the SSC team and allow the client to have confidence in the work being completed. My internship has allowed me to set and achieve many goals such as coming to understand that the field of cybersecurity truly is the right place for me. I have also learned that it’s ok to be wrong if you learn something from it. For significant accomplishments I have helped integrate standards mapping into the fabric of SSC. The field experience that I have gained such as interacting with the SSC team along with other senior staff and building a good rapport are crucial skills to. This time at PNNL has been an irreplaceable experience.

42 ENGINEERING↗

Environmental dependence of the galaxy stellar mass function in the Dark Energy Survey Science Verification Data

Measurements of the galaxy stellar mass function are crucial to understand the formation of galaxies in the Universe. In a hierarchical clustering paradigm it is plausible that there is a connection between the properties of galaxies and their environments. Evidence for environmental trends has been established in the local Universe. The Dark Energy Survey (DES) provides large photometric datasets that enable further investigation of the assembly of mass. In this study we use ~3.2 million galaxies from the (South Pole Telescope) SPT-East field in the DES science verification (SV) dataset. From grizY photometry we derive galaxy stellar masses and absolute magnitudes, and determine the errors on these properties using Monte-Carlo simulations using the full photometric redshift probability distributions. We compute galaxy environments using a fixed conical aperture for a range of scales. We construct galaxy environment probability distribution functions and investigate the dependence of the environment errors on the aperture parameters. We compute the environment components of the galaxy stellar mass function for the redshift range 0.15 < z < 1.05. For z < 0.75 we find that the fraction of massive galaxies is larger in high density environment than in low density environments. We show that the low density and high density components converge with increasing redshift up to z ~ 1.0 where the shapes of the mass function components are indistinguishable. As a result, our study shows how high density structures build up around massive galaxies through cosmic time.

79 ASTRONOMY AND ASTROPHYSICS↗

AI Benchmark Democratization and Carpentry

Benchmarks are a cornerstone of modern machine learning, enabling reproducibility, comparison, and scientific progress. However, AI benchmarks are increasingly complex, requiring dynamic, AI-focused workflows. Rapid evolution in model architectures, scale, datasets, and deployment contexts makes evaluation a moving target. Large language models often memorize static benchmarks, causing a gap between benchmark results and real-world performance. Beyond traditional static benchmarks, continuous adaptive benchmarking frameworks are needed to align scientific assessment with deployment risks. This calls for skills and education in AI Benchmark Carpentry. From our experience with MLCommons, educational initiatives, and programs like the DOE's Trillion Parameter Consortium, key barriers include high resource demands, limited access to specialized hardware, lack of benchmark design expertise, and uncertainty in relating results to application domains. Current benchmarks often emphasize peak performance on top-tier hardware, offering limited guidance for diverse, real-world scenarios. Benchmarking must become dynamic, incorporating evolving models, updated data, and heterogeneous platforms while maintaining transparency, reproducibility, and interpretability. Democratization requires both technical innovation and systematic education across levels, building sustained expertise in benchmark design and use. Benchmarks should support application-relevant comparisons, enabling informed, context-sensitive decisions. Dynamic, inclusive benchmarking will ensure evaluation keeps pace with AI evolution and supports responsible, reproducible, and accessible AI deployment. Community efforts can provide a foundation for AI Benchmark Carpentry.

von Laszewski, Gregor [Virginia U.]↗

Computer Aided Multi-scale Design of SiC-Si3N4 Nanoceramic Composites for High-Temperature Structural Applications

It is estimated that by using better and improved high temperature structural materials, the power generation efficiency of the power plants can be increased by 15% resulting in significant cost savings. One such promising material system for future high-temperature structural applications in power plants is Silicon Carbide-Silicon Nitride (SiC-Si{sub 3}N{sub 4}) nanoceramic matrix composites. The described research work focuses on multiscale simulation-based design of these SiC-Si{sub 3}N{sub 4} nanoceramic matrix composites. There were two primary objectives of the research: (1) Development of a multiscale simulation tool and corresponding multiscale analyses of the high-temperature creep and fracture resistance properties of the SiC-Si{sub 3}N{sub 4} nanocomposites at nano-, meso- and continuum length- and timescales; and (2) Development of a simulation-based robust design optimization methodology for application to the multiscale simulations to predict the range of the most suitable phase morphologies for the desired high-temperature properties of the SiC-Si{sub 3}N{sub 4} nanocomposites. The multiscale simulation tool is based on a combination of molecular dynamics (MD), cohesive finite element method (CFEM), and continuum level modeling for characterizing time-dependent material deformation behavior. The material simulation tool is incorporated in a variable fidelity model management based design optimization framework. Material modeling includes development of an experimental verification framework. Using material models based on multiscaling, it was found using molecular simulations that clustering of the SiC particles near Si{sub 3}N{sub 4} grain boundaries leads to significant nanocomposite strengthening and significant rise in fracture resistance. It was found that a control of grain boundary thicknesses by dispersing non-stoichiometric carbide or nitride phases can lead to reduction in strength however significant rise in fracture strength. The temperature dependent strength and microstructural stability was also significantly depended upon the dispersion of new phases at grain boundaries. The material design framework incorporates high temperature creep and mechanical strength data in order to develop a collaborative multiscale framework of morphology optimization. The work also incorporates a computer aided material design dataset development procedure where a systematic dataset on material properties and morphology correlation could be obtained depending upon a material processing scientist's requirements. Two different aspects covered under this requirement are: (1) performing morphology related analyses at the nanoscale and at the microscale to develop a multiscale material design and analyses capability; (2) linking material behavior analyses with the developed design tool to form a set of material design problems that illustrate the range of material design dataset development that could be performed. Overall, a software based methodology to design microstructure of particle based ceramic nanocomposites has been developed. This methodology has been shown to predict changes in phase morphologies required for achieving optimal balance of conflicting properties such as minimal creep strain rate and high fracture strength at high temperatures. The methodology incorporates complex material models including atomistic approaches. The methodology will be useful to design materials for high temperature applications including those of interest to DoE while significantly reducing cost of expensive experiments.

36 MATERIALS SCIENCE↗

Out of Distribution Detection with Neural Network Anchoring

This is code to reproduce and build on OOD detection from the paper "Out of Distribution Detection with Neural Network Anchoring". Our goal here is to exploit heteroscedastic temperature scaling as a calibration strategy for out of distribution (OOD) detection. Heteroscedasticity here refers to the fact that the optimal temperature parameter for each sample can be different, as opposed to conventional approaches that use the same value for the entire distribution. To enable this, we propose a new training strategy called anchoring that can estimate appropriate temperature values for each sample, leading to state-of-the-art OOD detection performance across several benchmarks. Using NTK theory, we show that this temperature function estimate is closely linked to the epistemic uncertainty of the classifier, which explains its behavior. In contrast to some of the best-performing OOD detection approaches, our method does not require exposure to additional outlier datasets, custom calibration objectives, or model ensembling. Through empirical studies with different OOD detection settings - far OOD, near OOD, and semantically coherent OOD - we establish a highly effective OOD detection approach.

Thiagarajan, Jayaraman↗