Search NASA⌕ Search

SEARCH · Search NASA

Results for “modulation classification”

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 91 records · Page 5
All sourcesNASA NTRSDOE OSTI
DOE OSTI2026

Microstructure prediction for Ti-22Al-25Nb in laser powder bed fusion

This work presents a physics-informed framework for predicting solidification morphology and defect susceptibility in additively manufactured Ti–22Al–25Nb across a broad processing space. The framework integrates solidification microstructure selection (SMS) analysis with a single-track defect-based printability map to establish a unified methodology linking processing parameters to both interfacial morphology and manufacturability. Thermal gradients G and solidification rates R are first computed using the Thermo-Calc Additive Manufacturing (TC-AM) module, a finite-interface-dissipation (FID) phase-field (PF) model coupled with CALPHAD method is then employed to systematically distinguish planar and dendritic regimes as functions of $G$ and $R$. By superimposing the printability map onto the morphology projections, a comprehensive process–structure framework is obtained. Across most processing conditions, the predicted microstructure is predominantly dendritic, while planar growth emerges only under selected laser power $P$ and scan speed $v$ combinations. In addition to morphology classification, the framework quantifies the dendritic area fraction and introduces a width-based morphology descriptor to characterize the spatial extent of planar/dendritic regions within the melt pool. It provides mechanistic insight into the interplay between solidification physics and defect formation, offering practical guidance for parameter selection and microstructural control in Ti–22Al–25Nb additive manufacturing (AM).

36 MATERIALS SCIENCE↗
NASA NTRS2008

Processing LiDAR Data to Predict Natural Hazards

ELF-Base and ELF-Hazards (wherein 'ELF' signifies 'Extract LiDAR Features' and 'LiDAR' signifies 'light detection and ranging') are developmental software modules for processing remote-sensing LiDAR data to identify past natural hazards (principally, landslides) and predict future ones. ELF-Base processes raw LiDAR data, including LiDAR intensity data that are often ignored in other software, to create digital terrain models (DTMs) and digital feature models (DFMs) with sub-meter accuracy. ELF-Hazards fuses raw LiDAR data, data from multispectral and hyperspectral optical images, and DTMs and DFMs generated by ELF-Base to generate hazard risk maps. Advanced algorithms in these software modules include line-enhancement and edge-detection algorithms, surface-characterization algorithms, and algorithms that implement innovative data-fusion techniques. The line-extraction and edge-detection algorithms enable users to locate such features as faults and landslide headwall scarps. Also implemented in this software are improved methodologies for identification and mapping of past landslide events by use of (1) accurate, ELF-derived surface characterizations and (2) three LiDAR/optical-data-fusion techniques: post-classification data fusion, maximum-likelihood estimation modeling, and hierarchical within-class discrimination. This software is expected to enable faster, more accurate forecasting of natural hazards than has previously been possible.

Fairweather, Ian↗
NASA NTRS1983

Accreting degenerate dwarfs in close binary systems

Advances in the study of cataclysmic variables made during the past few years are reviewed. The classification of cataclysmic binaries and their dynamic properties are summarized. The hard and soft X-ray emission from these objects is discussed, and two alternative accretion geometries for producing this radiation from deep in the potential well of the degenerate dwarf are considered. The ultraviolet and optical spectrum is addressed, including disk emission and contributions from the companion star. Magnetic fields in cataclysmic variables are discussed, and the temporal behavior of these stars is addressed, including periodic modulations associated with orbital motion and rotation as well as flickering and pulsation reflecting the mass transfer process and the dynamics of matter near the surface of the accreting star. The outburst process is considered, including classical novae, recurrent novae, and dwarf novae.

Cordova, F. A.↗
NASA NTRS2001

Microscale Particulate Classifiers (MiPAC) Being Developed

The NASA Glenn Research Center is developing microscale sensors to characterize atmospheric-borne particulates. The devices are fabricated using MEMS (microelectromechanical systems) technologies. These technologies are derived from those originally developed in support of the semiconductor processing industry. The resulting microsensors can characterize a wide range of particles and are, therefore, suitable to a broad range of applications. This project is supported under a collaborative program called the Glennan Microsystems Initiative. The initiative comprises members of NASA Glenn Research Center, various university affiliates from the State of Ohio, and a number of participating industrial partners. Funding is jointly provided by NASA, the State of Ohio, and industrial members. The work described here is a collaborative arrangement between researchers at Glenn, the University of Minnesota, The National Institute of Standards and Technology (NIST), and the Cleveland State University. Actual device fabrication is conducted at Glenn and at the laboratories of Case Western Reserve University. Case Western is also located in Cleveland, Ohio, and is a participating member of the initiative. The principal investigator for this project is Paul S. Greenberg of Glenn. Two basic types of devices are being developed, and target different ranges of particle sizes. The first class of devices, which is used to measure nanoparticles (i.e., particles in the range of 0.002 to 1 mm), is based on the technique of Electrical Mobility Classification. This technique also affords the valuable ability of measuring the electrical charge state of the particles. Such information is important in the understanding of agglomeration mechanisms and is useful in the development of methods for particle repulsion. The second type of device being developed, which utilizes optical scattering, is suitable for particles larger than 1 mm. This technique also provides information on particle shape and composition. Applications for these sensors include fundamental planetary climatology, monitoring and filtration in spacecraft, human habitation modules and related systems, characterization of particulate emissions from propulsion and power systems, and as early warning sensors for both space-based and ter-restrial fire detection. These devices are also suitable for characterizing biological compounds such as allergens, infectious agents, and biotoxic agents.

Greenberg, Paul S.↗
NASA NTRSDate not supplied

Developing Open-Source Training Materials for AI/ML and Space Biological Sciences Using NASA Cloud-Based Data

Artificial Intelligence (AI) and Machine Learning (ML) has gained significant traction in the biological and biomedical research fields in the last two decades, in part thanks to an increasing culture of open data sharing and reuse. Due to its capability for identifying complex relationships and patterns, AI/ML methodology is particularly well suited to recognize and predict biological patterns from high-dimensional next-generation sequencing data (e.g. whole genome sequencing, transcriptomic sequencing), as well as from biological or medical imaging data (e.g. microscopy, computed tomography, ultrasound, magnetic resonance imaging, radiography). These methodologies hold particular promise for space biosciences research and automated space health monitoring systems. However, there are many key considerations for properly training, validating, and testing a machine learning model in biological research or clinical application. Even with the positive culture of Open Science and data sharing, inexperienced researchers working quickly without proper checks can produce models that perform poorly outside of the immediate training dataset. Lessons learned from biological AI/ML research indicate that Open Science principles such as data sharing and open-source code must go hand-in-hand with publicly available, high-quality training curricula in best practices, with modules centered on real-life scientific use cases and data so future AI/ML practitioners gain experience on real problems. Here we present the development of open-source training materials for AI/ML and space biosciences, as part of the NASA Transform to Open Science Training (TOPST) initiative. We develop 4 independent training programs, focused on the following topics: 1) Fundamentals of Machine Learning and Space Biosciences Domain, 2) Open Science, Artificial Intelligence, and Ethical Best Practices for Data Sharing and Analysis, 3) Using AI/ML Classification to Identify Gene Networks Affected By Space Exposure in Mouse Liver, and 4) Using Neural Networks to Find DNA Damage Patterns in Immune Cells after Radiation. All programs leverage cloud-based NASA biological datasets. The curriculum we present will enable worldwide access to training in AI/ML and scientific analysis.

James Andrew Casaletto↗
NASA NTRSDate not supplied

Developing Open-Source Training Materials for AI/ML and Space Biological Sciences Using NASA Cloud-Based Data

Artificial Intelligence (AI) and Machine Learning (ML) has gained significant traction in the biological and biomedical research fields, in part due to a culture of open data sharing and reuse. AI/ML methodology is well-suited to recognize and predict biological patterns from high-dimensional next-generation sequencing data (e.g. whole genome sequencing, transcriptomic sequencing), as well as from biological or medical imaging data (e.g. microscopy, computed tomography, ultrasound, magnetic resonance imaging, radiography). These methodologies hold particular promise for space biosciences research and automated space health monitoring systems. However, there are key considerations for properly training, validating, and testing a machine learning model in biological research or clinical application. Inexperienced researchers can produce models that perform poorly outside of the training dataset. Open Science principles such as data sharing and open-source code must go hand-in-hand with publicly available, high-quality training curricula in best practices, with modules centered on real-life scientific use cases and data so future AI/ML practitioners gain experience on real problems. Here we present the development of open-source training materials for AI/ML and space biosciences, as part of the NASA Transform to Open Science Training (TOPST) initiative. We develop 4 independent training programs, focused on the following topics: 1) Fundamentals of Machine Learning and Space Biosciences Domain, 2) Open Science, Artificial Intelligence, and Ethical Best Practices for Data Sharing and Analysis, 3) Using AI/ML Classification to Identify Gene Networks Affected By Space Exposure in Mouse Liver, and 4) Using Neural Networks to Find DNA Damage Patterns in Immune Cells after Radiation. All programs leverage cloud-based NASA biological datasets. The curriculum we present will enable worldwide access to training in AI/ML and scientific analysis.

James Casaletto↗
NASA NTRSDate not supplied

The Ejectable Data Recorder: A Lean, Risk-Informed Approach for Hardware Development

NASA is developing the Orion spacecraft to transport crew from the Earth to the Moon as part of the Artemis series of missions. To provide a crew escape capability from pre-launch through ascent, the Orion vehicle is equipped with a Launch Abort System (LAS), built by Lockheed Martin, which pulls the capsule away from the launch vehicle in the event of an abort scenario. The Ascent Abort 2 (AA-2) test flight occurred on July 2, 2019,and tested a production version of the LAS to ensure that it can operate as intended, and to collect a large data set from hundreds of sensors on the vehicle to support Orion flight certification. In the original AA-2 architecture, a single-string set of communications antennas on the LAS would downlink all of the in-flight test data to ground stations. However, that communications architecture was predicted to have data dropouts during abort and jettison of the LAS, and would not support data transmission at all after LAS jettison. As a result, a comprehensive trade study was completed, yielding the addition of antennas on the crew module (CM), a buffer/rebroadcast capability for key portions of the flight, and an ejectable data recorder (EDR) subsystem. This EDR subsystem would serve as a backup to the radio frequency (RF) communications system, and would be non-flight critical, providing a unique capability that enabled management to take a different approach with the hardware and software development. The Crew Module and Separation Ring were developed as “Class 1”Flight Hardware, albeit with some tailoring approaches to enable efficiencies. The Class 1 designation requires full rigor for flight hardware and software, documenting everything that happens to a piece of hardware from procurement through disposal, requiring a full spectrum of acceptance tests, and the highest rigor of quality assurance processes. At the other end of the spectrum, Class 3hardware is controlled, but not intended for flight, and leaves the level of rigor up to the project manager. This classification is often used for research and development projects. Similarly,Class-1E has been recently defined at NASA for ISS payloads and technology development projects that are not flight critical and do not need the full rigor of Class 1 to be successful. The EDR subsystem was challenged at commencement to adopt a skunkworks and agile-like approach to hardware development, allowing for a different risk posture than the rest of the AA-2 hardware. After initially pursuing Class 1 processes, the EDR subsystem design evolved to incorporating numerous commercial components, leading to re-designation as a Class-1E subsystem. The resulting EDR subsystem was fully successful in meeting all flight system requirements, and achieved 100% retrieval of flight test data. This paper will discuss the risk posture of the EDR subsystem and the subsequent tailoring that was enacted as part of its Class-1E status.

EDR↗
DOE OSTI2025

Identification of shared viral sequences in peat moss metagenomes reveals elements of a possible Sphagnum core virome

Viruses are an understudied component of plant microbiomes. Identifying viruses that are shared between individual plants, or members of the “core virome”, could reveal stable viral populations with the potential to modulate the composition and function of the microbiome. Here, we examined the virome associated with Sphagnum mosses, a keystone species that has direct influence over the fate of peatland carbon stores. We analyzed bulk metagenomes and metatranscriptomes generated from Sphagnum field samples collected over a ten-month period to identify virus-like sequences shared among plants. Individual Sphagnum samples harbored distinct DNA and RNA viromes where only a small percentage (< 1%) of the total number of identified viral contigs were shared among all samples. Based on taxonomic classification, the shared viral contigs represent bacterial viruses, or phage (Caudoviricetes), as well as viruses of eukaryotes, namely nucleocytoplasmic large DNA viruses (Nucleocytoviricota) and RNA viruses (Riboviria). We linked the shared phage-like contigs to viral regions within sequenced genomes of bacterial taxa that are members of the Sphagnum core microbiome, suggesting that these contigs represent temperate phage or degraded prophage. The putative nucleocytoplasmic large DNA viruses and RNA viruses were phylogenetically diverse and showed sequence similarity to viruses associated with a broad range of hosts and environmental sources. The identification of shared viral contigs suggested that, despite the compositional heterogeneity between samples, Sphagnum mosses may harbor a core virome. Future work validating the presence of the core virome is warranted as it may aid in understanding how persistent viruses impact microbiome ecology and symbiont evolution within this climatically relevant keystone species.

Metagenomics↗
NASA NTRS2004

Experimental Test-Bed for Intelligent Passive Array Research

This document describes the test-bed designed for the investigation of passive direction finding, recognition, and classification of speech and sound sources using sensor arrays. The test-bed forms the experimental basis of the Intelligent Small-Scale Spatial Direction Finder (ISS-SDF) project, aimed at furthering digital signal processing and intelligent sensor capabilities of sensor array technology in applications such as rocket engine diagnostics, sensor health prognostics, and structural anomaly detection. This form of intelligent sensor technology has potential for significant impact on NASA exploration, earth science and propulsion test capabilities. The test-bed consists of microphone arrays, power and signal distribution modules, web-based data acquisition, wireless Ethernet, modeling, simulation and visualization software tools. The Acoustic Sensor Array Modeler I (ASAM I) is used for studying steering capabilities of acoustic arrays and testing DSP techniques. Spatial sound distribution visualization is modeled using the Acoustic Sphere Analysis and Visualization (ASAV-I) tool.

Solano, Wanda M.↗
NASA NTRS1993

An investigative study of multispectral data compression for remotely-sensed images using vector quantization and difference-mapped shift-coding

A study is conducted to investigate the effects and advantages of data compression techniques on multispectral imagery data acquired by NASA's airborne scanners at the Stennis Space Center. The first technique used was vector quantization. The vector is defined in the multispectral imagery context as an array of pixels from the same location from each channel. The error obtained in substituting the reconstructed images for the original set is compared for different compression ratios. Also, the eigenvalues of the covariance matrix obtained from the reconstructed data set are compared with the eigenvalues of the original set. The effects of varying the size of the vector codebook on the quality of the compression and on subsequent classification are also presented. The output data from the Vector Quantization algorithm was further compressed by a lossless technique called Difference-mapped Shift-extended Huffman coding. The overall compression for 7 channels of data acquired by the Calibrated Airborne Multispectral Scanner (CAMS), with an RMS error of 15.8 pixels was 195:1 (0.41 bpp) and with an RMS error of 3.6 pixels was 18:1 (.447 bpp). The algorithms were implemented in software and interfaced with the help of dedicated image processing boards to an 80386 PC compatible computer. Modules were developed for the task of image compression and image analysis. Also, supporting software to perform image processing for visual display and interpretation of the compressed/classified images was developed.

Jaggi, S.↗
DOE OSTI2024

Chemical reaction enhanced graph learning for molecule representation

Abstract Motivation Molecular representation learning (MRL) models molecules with low-dimensional vectors to support biological and chemical applications. Current methods primarily rely on intrinsic molecular information to learn molecular representations, but they often overlook effectively integrating domain knowledge into MRL. Results In this article, we develop a reaction-enhanced graph learning (RXGL) framework for MRL, utilizing chemical reactions as domain knowledge. RXGL introduces dual graph learning modules to model molecule representation. One module employs graph convolutions on molecular graphs to capture molecule structures. The other module constructs a reaction-aware graph from chemical reactions and designs a novel graph attention network on this graph to integrate reaction-level relations into molecular modeling. To refine molecule representations, we design a reaction-based relation learning task, which considers the relations between the reactant and product sides in reactions. In addition, we introduce a cross-view contrastive task to strengthen the cooperative associations between molecular and reaction-aware graph learning. Experiment results show that our RXGL achieves strong performance in various downstream tasks, including product prediction, reaction classification, and molecular property prediction. Availability and implementation The code is publicly available at https://github.com/coder-ACAC/RLM.

Biochemistry & Molecular Biology↗
DOE OSTI2025

United States multi-sector land use and land cover base maps to support human and Earth system models

Abstract Earth System Models (ESMs) require current and future projections of land use and landcover change (LULC) to simulate land-atmospheric interactions and global biogeochemical cycles. Among the most utilized land systems in ESMs are the Community Land Model (CLM) and the Land-Use Harmonization 2 (LUH2) products. Regional studies also use these products by extending coarse projections to finer resolutions via downscaling or by using multisector dynamic (MSD) models. One such MSD model is the Global Change Analysis Model (GCAM), which has its own independent land module, but often relies on CLM or LUH2 as spatial inputs for its base years. However, this requires harmonization of thematically incongruent land systems at multiple spatial resolutions, leading to uncertainty and error propagation. To resolve these issues, we develop a thematically consistent LULC system for the conterminous United States adaptable to multiple MSD frameworks to support research at a regional level. Using empirically derived spatial products, we developed a series of base maps for multiple contemporary years of observation at a 30-m resolution that support flexibility and interchangeability amongst LUH2, CLM, and GCAM classification systems.

Oliver, Jay↗
NASA NTRS1980

Optical identification of 4U1907 + 09 using the HEAO-1 scanning modulation collimator position

The paper reports an optical identification of 4U1907 + 09 with a m(v) = 16.4 stellar object in the location determined by the scanning modulation collimator experiment on the first High Energy Astronomy Observatory (HEAO-1). The identification is based on the presence of very strong and broad H-alpha emission. The optical data constrain the distance to be 2-13 kpc, and this gives a range of uncertainty to the typical 2-10 keV luminosity of (1 x 10 to the 35th to 3 x 10 to the 36th) erg/s. The X-ray spectrum and variability is reported, and consideration is given to the hypothesis that the object is an OB supergiant, although its faintness in the blue makes precise spectral classification impossible. It is suggested that this system is an example of a luminous, massive primary emitting a stellar wind which is accreted on the compact object.

Schwartz, D. A.↗
NASA NTRS1991

Three Hierarchies in Skeletal Muscle Fibre Classification Allotype, Isotype and Phenotype

Immunocytochemical analyses using specific anti-myosin antibodies of mammalian muscle fibers during regeneration, development, and after denervation have revealed two distinct myogenic components determining fiber phenotype. The jaw-closing muscles of the cat contain superfast fibers which express a unique myosin not found in limb muscles. When superfast muscle is transplanted into a limb muscle bed, regenerating myotubes synthesize superfast myosin independent of innervation. Reinnervation by the nerve to a fast muscle leads to the expression of superfast and not fast myosin, while reinnervation by the nerve to a slow muscle leads to the expression of a slow myosin. When limb muscle is transplanted into the jaw muscle bed, only limb myosins are synthesized. Thus jaw and limb muscles belong to distinct allotypes, each with a unique range of phenotype options, the expressions of which may be modulated by the nerve. Primary and secondary myotubes in developing jaw and limb muscles are observed to belong to different categories characterized by different patterns of myosin gene expression. By taking into consideration the pattern of myosins synthesized and the changes in fiber size after denervation, 3 types of primary (fast, slow, and intermediate) fibers can be distinguished in rat fast limb muscles. All primaries synthesize slow myosin soon after their formation, but this is withdrawn in fast and intermediate primaries at different times. After neonatal denervation, slow and intermediate primaries express slow primaries hypertrophy with other fibers atrophy. In the mature rat, the number of slow fibers in the EDL is less than the number of slow primaries. Upon denervation, hypertrophic slow fibers matching the number and topographic distribution of slow primaries appear, suggesting that a subpopulation of the slow primaries acquire the fast phenotype during adult life, but reveal their original identity as slow primaries in response to denervation by hypertrophying and synthesizing slow myosin. It is proposed that within each muscle allotype, the various isotypes of primary and secondary fibers are myogenically determined, and are derived from different lineage of myoblasts.

Hoh, Joseph F. Y.↗
DOE OSTI2025

Unrolled Video Super-Resolution Network with Autoregressive Prior for the Case of Known Motion

Real-time detection and classification of distant objects is necessary for many national security applications. However, when objects are far from the sensor, they occupy only a small number of pixels in the captured video, limiting the amount of visual detail available for recognition. State-of-the-art classification methods typically rely on high-resolution (HR) video streams to capture characteristic object features, but obtaining such detail is challenging for distant objects that occupy only a few pixels. This motivates the development of video super-resolution (VSR) methods that enhance object classification by recovering fine details from low-pixel representations. Current VSR methods rely either on model-based optimization, which is interpretable but computationally expensive, or on learning-based approaches, which are efficient and high-performing but often lack flexibility and interpretability. In this report, we propose an end-to-end trainable unrolled VSR network, UVSRNet, which super-resolves each frame in a video by exploiting sub-pixel motion between neighboring low-resolution (LR) frames as well as incorporating high-frequency detail from previously super-resolved frames. In particular, by unrolling a plug-and-play (PnP) half-quadratic splitting (HQS) algorithm, we leverage a model-based data-fitting module alongside a learning-based autoregressive prior module. This combination yields a method that maintains the flexibility and interpretability of model-based methods while achieving the performance advantages of learning-based methods.

45 MILITARY TECHNOLOGY, WEAPONRY, AND NATIONAL DEF↗
NASA NTRS2018

A NICER Look at the Aql X-1 Hard State

We report on a spectral-timing analysis of the neutron star low-mass X-ray binary(LMXB)AqlX-1 with the Neutron Star Interior Composition Explorer (NICER) on the International Space Station (ISS). AqlX-1 wasobserved with NICER during a dim outburst in 2017 July, collecting approximately 50 ks of good exposure. The spectral and timing properties of the source correspond to that of a (hard) extreme island state in the atoll classification. We find that the fractional amplitude of the low-frequency (<0.3Hz) band-limited noise shows adramatic turnover as a function of energy: it peaks at 0.5keV with nearly 25% rms, drops to 12% rms at 2keV,and rises to 15% rms at 10keV. Through the analysis of covariance spectra, we demonstrate that band-limited noise exists in both the soft thermal emission and the power-law emission. Additionally, we measure hard timelags, indicating the thermal emission at 0.5keV leads the power-law emission at 10 keV on a timescale of 100ms at 0.3Hz to10ms at 3Hz. Our results demonstrate that the thermal emission in the hard state is intrinsically variable, and is driving the modulation of the higher energy power-law. Interpreting the thermal spectrum as disk emission, we find that our results are consistent with the disk propagation model proposed for accretion onto black holes.

Bult, Peter↗
NASA NTRS1985

New identifications of bright X-ray sources with the HEAO-1 Scanning Modulation Collimator

Based on data obtained with the HEAO-1 Scanning Modulation Collimator (MC) experiment, candidate Be star systems and BL Lac objects are reported. Identification of the V = 6.6 star HD91188 as a Be star is confirmed, and X-ray luminosities ranging from 9 x 10 to the 32nd to 1.4 x 10 to the 34th ergs/s are derived. A B2 spectral type with a reddening E(B - V) = 0.42 is deduced for a V = 9.87 star which appears in an MC location diamond for the source 1H0550 + 286. A V = 16.5 ultraviolet excess object in the location 2A1058 - 226 = 3A1057 - 224 = 4U1057 - 21 = H1100 - 230 gave a 20 cm flux of 83 mJy, and a V = 16.2 ultraviolet excess object at 3A1422 + 425 = 1H143 + 423 was found to have a 20 cm flux of 33 mJy, suggesting their classification as BL Lac objects.

Schwartz, D. A.↗
DOE OSTI2026

Maximizing dynamic range and performance of anatase TiO 2 ECRAM through structure and programming

Here, in this study, we investigate the structure-dependent modulation characteristics of all-solid-state three-terminal electrochemical random-access memory (ECRAM) based on an anatase Li x TiO 2 channel. By directly comparing “asymmetric” and “symmetric” ECRAM device architectures, we reveal significant insight into the impact of a non-zero gate-drain open-circuit voltage and its influence on voltage vs. current-controlled gating. We also explore the impact of potentiation/depression write parameters on the symmetry, linearity, and dynamic range of the device response. Together, initial results from optimizing structure and programming approaches yielded unprecedented G max /G min ratios of >1,000 for ECRAM and hundreds of tunable memory states with excellent linearity and symmetry. Simulations based on these ECRAM devices further illustrate the promise of this analog memory technology, achieving near 2% classification error in the MNIST digit recognition benchmark for a range of training parameters compared to a theoretical best of 1.66% and outperforming other device models extracted from the literature.

AIHWKit↗