Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Error Mitigation”

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 721 records · Page 40

Impact of Pilot Delay and Non-Responsiveness on the Safety Performance of Airborne Separation

Assessing the safety effects of prediction errors and uncertainty on automationsupported functions in the Next Generation Air Transportation System concept of operations is of foremost importance, particularly safety critical functions such as separation that involve human decision-making. Both ground-based and airborne, the automation of separation functions must be designed to account for, and mitigate the impact of, information uncertainty and varying human response. This paper describes an experiment that addresses the potential impact of operator delay when interacting with separation support systems. In this study, we evaluated an airborne separation capability operated by a simulated pilot. The experimental runs are part of the Safety Performance of Airborne Separation (SPAS) experiment suite that examines the safety implications of prediction errors and system uncertainties on airborne separation assistance systems. Pilot actions required by the airborne separation automation to resolve traffic conflicts were delayed within a wide range, varying from five to 240 seconds while a percentage of randomly selected pilots were programmed to completely miss the conflict alerts and therefore take no action. Results indicate that the strategicAirborne Separation Assistance System (ASAS) functions exercised in the experiment can sustain pilot response delays of up to 90 seconds and more, depending on the traffic density. However, when pilots or operators fail to respond to conflict alerts the safety effects are substantial, particularly at higher traffic densities.

Consiglio, Maria↗

Addressing bias in bagging and boosting regression models

As artificial intelligence (AI) becomes widespread, there is increasing attention on investigating bias in machine learning (ML) models. Previous research concentrated on classification problems, with little emphasis on regression models. This paper presents an easy-to-apply and effective methodology for mitigating bias in bagging and boosting regression models, that is also applicable to any model trained through minimizing a differentiable loss function. Our methodology measures bias rigorously and extends the ML model's loss function with a regularization term to penalize high correlations between model errors and protected attributes. We applied our approach to three popular tree-based ensemble models: a random forest model (RF), a gradient-boosted model (GBT), and an extreme gradient boosting model (XGBoost). We implemented our methodology on a case study for predicting road-level traffic volume, where RF, GBT, and XGBoost models were shown to have high accuracy. Despite high accuracy, the ML models were shown to perform poorly on roads in minority-populated areas. Our bias mitigation approach reduced minority-related bias by over 50%.

97 MATHEMATICS AND COMPUTING↗

Robust Damage-Mitigating Control of Aircraft for High Performance and Structural Durability

This paper presents the concept and a design methodology for robust damage-mitigating control (DMC) of aircraft. The goal of DMC is to simultaneously achieve high performance and structural durability. The controller design procedure involves consideration of damage at critical points of the structure, as well as the performance requirements of the aircraft. An aeroelastic model of the wings has been formulated and is incorporated into a nonlinear rigid-body model of aircraft flight-dynamics. Robust damage-mitigating controllers are then designed using the H(infinity)-based structured singular value (mu) synthesis method based on a linearized model of the aircraft. In addition to penalizing the error between the ideal performance and the actual performance of the aircraft, frequency-dependent weights are placed on the strain amplitude at the root of each wing. Using each controller in turn, the control system is put through an identical sequence of maneuvers, and the resulting (varying amplitude cyclic) stress profiles are analyzed using a fatigue crack growth model that incorporates the effects of stress overload. Comparisons are made to determine the impact of different weights on the resulting fatigue crack damage in the wings. The results of simulation experiments show significant savings in fatigue life of the wings while retaining the dynamic performance of the aircraft.

Caplin, Jeffrey↗

Ultracoherent SRF Cavity-Based Multi-Qudit Platform with Error-Resilient Control

Superconducting radio-frequency (SRF) cavities offer a promising platform for quantum computing due to their long coherence times, yet integrating nonlinear elements like transmons for control often introduces additional loss. We report a multimode quantum system based on a 2-cell elliptical shaped SRF cavity, comprising two cavity modes weakly coupled to an ancillary transmon circuit, designed to preserve coherence while enabling efficient control of the cavity modes. We mitigate the detrimental effects of the transmon decoherence through careful design optimization that reduces transmon-cavity couplings and participation in the dielectric substrate and lossy interfaces, to achieve single-photon lifetimes of 20.6\,ms and 15.6\,ms for the two modes, and a pure dephasing time exceeding 40\,ms. This marks an order-of-magnitude improvement over prior 3D multimode memories. Leveraging sideband interactions and novel error-resilient protocols, including measurement-based correction and post-selection, we achieve high-fidelity control over quantum states. This enables the preparation of Fock states up to $N = 20$ with fidelities exceeding 95\%, the highest reported to date to the authors' knowledge, as well as two-mode entanglement with an estimated coherence-limited fidelities of 99.9\% after post-selection. These results establish our platform as a robust foundation for quantum information processing, allowing for future extensions to high-dimensional qudit encodings.

Kim, T. [Northwestern U.]↗

Ultracoherent superconducting cavity-based multiqudit platform with error-resilient control

Superconducting radio-frequency (SRF) cavities offer a promising platform for quantum computing due to their long coherence times, yet integrating nonlinear elements like transmons for control often introduces additional loss. We report a multimode quantum system based on a 2-cell elliptical shaped SRF cavity, comprising two cavity modes weakly coupled to an ancillary transmon circuit, designed to preserve coherence while enabling efficient control of the cavity modes. We mitigate the detrimental effects of the transmon decoherence through careful design optimization that reduces transmon-cavity couplings and participation in the dielectric substrate and lossy interfaces, to achieve single-photon lifetimes of 20.6 ms and 15.6 ms for the two modes, and a pure dephasing time exceeding 40 ms. This marks an order-of-magnitude improvement over prior 3D multimode memories. Leveraging sideband interactions and novel error-resilient protocols, including measurement-based correction and post-selection, we achieve high-fidelity control over quantum states. This enables the preparation of Fock states up to $N = 20$ with fidelities exceeding 95%, the highest reported to date to the authors' knowledge, as well as two-mode entanglement with an estimated coherence-limited fidelities of 99.9% after post-selection. These results establish our platform as a robust foundation for quantum information processing, allowing for future extensions to high-dimensional qudit encodings.

Kim, Taeyoon [Fermilab; Northwestern U.]↗

Solid State Transformer Controls for Mitigation of E3a High-Altitude Electromagnetic Pulse Insults

This paper explores the use of a solid state transformer (SST) to mitigate the 𝐸 3𝐴 component of a high-altitude electromagnetic pulse (HEMP) insult using external energy storage optimal control techniques. In lieu of conventional passive blocking devices or feedback-controlled energy storage devices, a novel implementation of Hamiltonian error tracking is utilized to develop a feedback control law for the variable converter ratio in an SST. The findings of the simulations performed in this paper suggest that additional energy storage is not necessary to protect an individual load from a HEMP insult. The simulations performed examine the response of a single-phase SST connected to a single voltage source on a long transmission line on the one side and a single linear resistor on the other. The control law is specifically developed for the late-time, low-frequency portion of a HEMP insult, namely the 𝐸 3𝐴 components. The Hamiltonian error-based converter ratio control law is compared with nonlinear optimal feedforward controls to show that the HSSPFC is an external energy storage optimal controller.

HEMP mitigation↗

Ultracoherent superconducting cavity-based multiqudit platform with error-resilient control

Superconducting radio-frequency (SRF) cavities offer a promising platform for quantum computing due to their long coherence times, yet integrating nonlinear elements like transmons for control often introduces additional loss. We report a multimode quantum system based on a 2-cell elliptical-shaped SRF cavity, comprising two cavity modes weakly coupled to an ancillary transmon circuit, designed to preserve coherence while enabling efficient control of the cavity modes. We mitigate the detrimental effects of the transmon decoherence through careful design optimization that reduces transmon-cavity couplings and participation in the dielectric substrate and lossy interfaces, to achieve single-photon lifetimes of 20.6 ms and 15.6 ms for the two modes, and a pure dephasing time exceeding 40 ms. This marks an order-of-magnitude improvement over prior 3D multimode memories. Leveraging sideband interactions and novel error-resilient protocols, including measurement-based correction and post-selection, we achieve high-fidelity control over quantum states. This enables the preparation of Fock states up to N = 20 with fidelities exceeding 95%, the highest reported to date to the authors' knowledge, as well as two-mode entanglement with an estimated coherence-limited fidelities of 99.9% after post-selection. These results establish our platform as a robust foundation for quantum information processing, allowing for future extensions to high-dimensional qudit encodings.

Lu, Yao [Fermilab] (ORCID:000000020413698X)↗

Ultracoherent superconducting cavity-based multiqudit platform with error-resilient control

Superconducting radio-frequency (SRF) cavities offer a promising platform for quantum computing due to their long coherence times, yet integrating nonlinear elements like transmons for control often introduces additional loss. We report a multimode quantum system based on a 2-cell elliptical-shaped SRF cavity, comprising two cavity modes weakly coupled to an ancillary transmon circuit, designed to preserve coherence while enabling efficient control of the cavity modes. We mitigate the detrimental effects of the transmon decoherence through careful design optimization that reduces transmon-cavity couplings and participation in the dielectric substrate and lossy interfaces, to achieve single-photon lifetimes of 20.6 ms and 15.6 ms for the two modes, and a pure dephasing time exceeding 40 ms. This marks an order-of-magnitude improvement over prior 3D multimode memories. Leveraging sideband interactions and novel error-resilient protocols, including measurement-based correction and post-selection, we achieve high-fidelity control over quantum states. This enables the preparation of Fock states up to N = 20 with fidelities exceeding 95%, the highest reported to date to the authors' knowledge, as well as two-mode entanglement with an estimated coherence-limited fidelities of 99.9% after post-selection. These results establish our platform as a robust foundation for quantum information processing, allowing for future extensions to high-dimensional qudit encodings.

Lu, Yao [Fermilab] (ORCID:000000020413698X)↗

Implications of the fractional charge of hydroxide at the electrochemical interface

Rational design of materials that efficiently convert electrical energy into chemical bonds will ultimately depend on a thorough understanding of the electrochemical interface at the atomic level. Towards this goal, the use of density functional theory (DFT) at the generalized gradient approximation (GGA) level has been applied widely in the past 15 years. In the calculation of electrochemical reaction energetics using GGA-DFT, it is frequently implicitly assumed that ions in the Helmholtz plane have unit charge. However, the ion charge is observed to be fractional near the interface through both a capacitor model and through Bader charge partitioning. In this work, we show that this spurious charge transfer can be effectively mitigated by continuum charging of the electrolyte. Finally, we show that, similar to hydronium, the observed fractional charge of hydroxide is not due to a GGA level self-interaction error, as the partial charge is observed even when using hybrid level exchange–correlation functionals.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Novel Technique for Developing Linearized Convex System Models from Experimentally Derived Data

This paper presents a novel technique for generating a convex system model from an experimentally derived data set which features variance among repeated measurements. The convex system model developed as a test case characterizes the dynamic system losses of a vanadium redox flow battery as a function of the active power output and the battery state of charge. The technique hinges on a pre-cleaning via clustering procedure which precedes the formation of a planar convex hull comprised of triangular simplices. The clustering procedure efficiently reduces the experimental data set while mitigating variance among repeated measurements and removes outliers. Ultimately, the lower evelope of the planar convex hull serves as the desired convex system model. The proposed technique reduces systematic model error which is otherwise present when directly developing a planar convex hull model based on an unreduced data set.

24 POWER TRANSMISSION AND DISTRIBUTION↗

TU-FG-201-11: Evaluating the Validity of Prospective Risk Analysis Methods: A Comparison of Traditional FMEA and Modified Healthcare FMEA

Purpose: To examine the ability of traditional Failure mode and effects analysis (FMEA) and a light version of Healthcare FMEA (HFMEA), called Scenario analysis of FMEA (SAFER) by comparing their outputs in terms of the risks identified and their severity rankings. Methods: We applied two prospective methods of the quality management to surface image guided, linac-based radiosurgery (SIG-RS). For the traditional FMEA, decisions on how to improve an operation are based on risk priority number (RPN). RPN is a product of three indices: occurrence, severity and detectability. The SAFER approach; utilized two indices-frequency and severity-which were defined by a multidisciplinary team. A criticality matrix was divided into 4 categories; very low, low, high and very high. For high risk events, an additional evaluation was performed. Based upon the criticality of the process, it was decided if additional safety measures were needed and what they comprise. Results: Two methods were independently compared to determine if the results and rated risks were matching or not. Our results showed an agreement of 67% between FMEA and SAFER approaches for the 15 riskiest SIG-specific failure modes. The main differences between the two approaches were the distribution of the values and the failure modes (No.52, 54, 154) that have high SAFER scores do not necessarily have high FMEA RPN scores. In our results, there were additional risks identified by both methods with little correspondence. In the SAFER, when the risk score is determined, the basis of the established decision tree or the failure mode should be more investigated. Conclusion: The FMEA method takes into account the probability that an error passes without being detected. SAFER is inductive because it requires the identification of the consequences from causes, and semi-quantitative since it allow the prioritization of risks and mitigation measures, and thus is perfectly applicable to clinical parts of radiotherapy.

60 APPLIED LIFE SCIENCES↗

Validation and Scaling of Soil Moisture in a Semi-Arid Environment: SMAP Validation Experiment 2015 (SMAPVEX15)

The NASA SMAP (Soil Moisture Active Passive) mission conducted the SMAP Validation Experiment 2015 (SMAPVEX15) in order to support the calibration and validation activities of SMAP soil moisture data products. The main goals of the experiment were to address issues regarding the spatial disaggregation methodologies for improvement of soil moisture products and validation of the in situ measurement upscaling techniques. To support these objectives high-resolution soil moisture maps were acquired with the airborne PALS (Passive Active L-band Sensor) instrument over an area in southeast Arizona that includes the Walnut Gulch Experimental Watershed (WGEW), and intensive ground sampling was carried out to augment the permanent in situ instrumentation. The objective of the paper was to establish the correspondence and relationship between the highly heterogeneous spatial distribution of soil moisture on the ground and the coarse resolution radiometer-based soil moisture retrievals of SMAP. The high-resolution mapping conducted with PALS provided the required connection between the in situ measurements and SMAP retrievals. The in situ measurements were used to validate the PALS soil moisture acquired at 1-km resolution. Based on the information from a dense network of rain gauges in the study area, the in situ soil moisture measurements did not capture all the precipitation events accurately. That is, the PALS and SMAP soil moisture estimates responded to precipitation events detected by rain gauges, which were in some cases not detected by the in situ soil moisture sensors. It was also concluded that the spatial distribution of the soil moisture resulted from the relatively small spatial extents of the typical convective storms in this region was not completely captured with the in situ stations. After removing those cases (approximately10 of the observations) the following metrics were obtained: RMSD (root mean square difference) of0.016m3m3 and correlation of 0.83. The PALS soil moisture was also compared to SMAP and in situ soil moisture at the 36-km scale, which is the SMAP grid size for the standard product. PALS and SMAP soil moistures were found to be very similar owing to the close match of the brightness temperature measurements and the use of a common soil moisture retrieval algorithm. Spatial heterogeneity, which was identified using the high-resolution PALS soil moisture and the intensive ground sampling, also contributed to differences between the soil moisture estimates. In general, discrepancies found between the L-band soil moisture estimates and the 5-cm depth in situ measurements require methodologies to mitigate the impact on their interpretations in soil moisture validation and algorithm development. Specifically, the metrics computed for the SMAP radiometer-based soil moisture product over WGEW will include errors resulting from rainfall, particularly during the monsoon season when the spatial distribution of soil moisture is especially heterogeneous.

SMAPVEX15↗

Mitigating Photon Jitter in Optical PPM Communication

A theoretical analysis of photon-arrival jitter in an optical pulse-position-modulation (PPM) communication channel has been performed, and now constitutes the basis of a methodology for designing receivers to compensate so that errors attributable to photon-arrival jitter would be minimized or nearly minimized. Photon-arrival jitter is an uncertainty in the estimated time of arrival of a photon relative to the boundaries of a PPM time slot. Photon-arrival jitter is attributable to two main causes: (1) receiver synchronization error [error in the receiver operation of partitioning time into PPM slots] and (2) random delay between the time of arrival of a photon at a detector and the generation, by the detector circuitry, of a pulse in response to the photon. For channels with sufficiently long time slots, photon-arrival jitter is negligible. However, as durations of PPM time slots are reduced in efforts to increase throughputs of optical PPM communication channels, photon-arrival jitter becomes a significant source of error, leading to significant degradation of performance if not taken into account in design. For the purpose of the analysis, a receiver was assumed to operate in a photon- starved regime, in which photon counts follow a Poisson distribution. The analysis included derivation of exact equations for symbol likelihoods in the presence of photon-arrival jitter. These equations describe what is well known in the art as a matched filter for a channel containing Gaussian noise. These equations would yield an optimum receiver if they could be implemented in practice. Because the exact equations may be too complex to implement in practice, approximations that would yield suboptimal receivers were also derived.

Moision, Bruce↗

A Risk Analysis Tool for Estimating the Risk of Electrical Failures Due to Human Induced Defects

Aerospace electrical systems are required to withstand and adequately operate in extremely harsh environments that include, for example, high radiation exposure, temperature extremes, intense vibrational stress and drastic temperature cycling. The nature of aerospace electronics also demands high reliability since, with very few exceptions, there is no chance for hardware servicing or repairs. Common risk mitigation techniques for this type of situation are to perform a Reliability Analysis of the system throughout the development cycle, and to use electrical components that are regarded as “high reliability” because of additional controls and requirements applied in their design, manufacturing and testing. Unfortunately, studies have shown that even though these techniques are used, many systems fail to meet mission requirements well before the predicted lifetimes. This paper presents the analysis of failures of electrical parts, experienced during various stages of system development, at NASA Goddard Space Flight Center, Greenbelt MD, between the years 2001 and 2013. These components were subjected to qualification, screening and testing in which the goal was to ensure that the components would survive the stresses of the mission. The analysis categorizes failures by part type and failure mechanisms. One of the results of the analysis was the realization that a surprising proportion of failures experienced during system integration and testing were caused by human error (i.e. human induced defect). Further analysis included the determination of root failure mechanisms and any influencing factors contributing to these failures. The major causes of these defects were attributed to electrostatic damage (ESD), electrical overstress (EOS), mechanical overstress (MOS), and thermal overstress (TOS). Finally, the study proposes a risk analysis tool which incorporates these major causes for the failures, termed error-producing conditions (EPCs), and a proportionality factor representing the number of each type of failure that has occurred at the facility under study. These factors are quantified and used to communicate the risk of human induced defects for the assembly, integration and testing of space hardware based on the system’s electrical parts list. The new risk identification can trigger risk-mitigating actions more effectively, based on the presence of component categories or other hazardous conditions that have a history of failure due to human error.

Majewicz, Peter J.↗

Heat Transfer Salts for Nuclear Reactor Systems - Chemistry Control, Corrosion Mitigation, and Modeling

The concept of a molten salt reactor has existed for nearly sixty years. Previously all work was done during a large collaborative effort at Oak Ridge National Laboratory, culminating in a research reactor which operated for 15,000 hours without major error. This technical success has garnished interest in modern, high temperature, reactor schemes. Research using molten fluoride salts for nuclear applications requires a steady supply of high grade molten salts. There is no bulk supplier of research grade fluoride salts in the world, so a facility which could provide all the salt needed for testing at the University of Wisconsin had to be produced. Two salt purification devices were made for this purpose, a large scale purifier, and a small scale purifier, each designed to clean the salts from impurities and reduce their corrosion potential. As of now, the small scale has performed with flibe salt, hydrogen, and hydrogen fluoride, yielding clean salt. This salt is currently being used in corrosion testing facilities at the Massachusetts Institute of Technology and the University of Wisconsin. Working with the beryllium based salts requires extensive safety measures and health monitoring to prevent the development of acute or chronic beryllium disease, two pulmonary diseases created by an allergic reaction to beryllium in the lungs. Extensive health monitoring, engineering controls, and environment monitoring had to be set up with the University of Wisconsin department of Environment, Health and Safety. The hydrogen fluoride required for purification was also an extreme health hazard requiring thoughtful planning and execution. These dangers have made research a slow and tedious process. Simple processes, such as chemical handling and clean-up, can take large amounts of ingenuity and time. Other work has complemented the experimental research at Wisconsin to advance high temperature reactor goals. Modeling work has been performed in house to re-evaluate thermophysical properties of flibe and flinak. Pacific Northwest National Laboratories has focused on evaluating the fluorinating gas nitrogen trifluoride as a potential salt purification agent. Work there was performed on removing hydroxides and oxides from flinak salt under controlled conditions. Lastly, the University of California Berkeley has spent considerable time designing and simulating reactor components with fluoride salts at high temperatures. Despite the hurdles presented by the innate chemical hazards, considerable progress has been made. The stage has been set to perform new research on salt chemical control which could advance the fluoride salt cooled reactor concept towards commercialization. What were previously thought of as chemical undesirable, but nuclear certified, alloys have been shown to be theoretically compatible with fluoride salts at high temperatures. This preliminary report has been prepared to communicate the construction of the basic infrastructure required for flibe, as well as suggest original research to performed at the University of Wisconsin. Simultaneously, the contents of this report can serve as a detailed, but introductory guide to allow anyone to learn the fundamentals of chemistry, engineering, and safety required to work with flibe salt.

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗

Systematic and statistical uncertainties of the hilbert-transform based high-precision FID frequency extraction method

Pulsed nuclear magnetic resonance (NMR) is widely used in high-precision magnetic field measurements. The absolute value of the magnetic field is determined from the precession frequency of nuclear magnetic moments. The Hilbert transform is widely used to extract the phase function from the observed free induction decay (FID) signal and then its frequency. In this paper, a detailed implementation of a Hilbert-transform based FID frequency extraction method is described. How artifacts and noise level in the FID signal affect the extracted phase function are derived analytically. A method of mitigating the artifacts in the extracted phase function of an FID is discussed. Correlations between noises of the phase function samples are studied for different noise spectra. We discovered that the error covariance matrix for the extracted phase function is nearly singular and improper for constructing the $\chi^2$ used in the fitting routine. A down-sampling method for fixing the singular covariance matrix has been developed, so that the minimum $\chi^2$-fit yields properly the statistical uncertainty of the extracted frequency. Other practical methods of obtaining the statistical uncertainty are also discussed.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Impacts of hybridization and forecast errors on the probabilistic capacity credit of batteries

Battery storage is increasingly identified as being among the least-cost mix of technologies in the evolving U.S. electricity mix. This study explores the marginal capacity credit of batteries using a probabilistic, reliability-based, effective firm capacity method, which we apply for multiple battery power ratings, durations, coupling types, deployment locations, and dispatch profiles within a test system that is based on the Texas Interconnection in the year 2024. We find that the capacity credits for all battery durations depend on their ability to predict the timing of reliability events. Even 1-2 h forecast errors - resulting in early or delayed battery discharging relative to the onset of a reliability event - lead pronounced capacity credit reductions, especially for 4-h duration batteries. Coupling batteries with solar mitigates the uncertainty associated with a shorter-duration battery's availability during reliability events, primarily due to the relatively high solar capacity credit in our test system. Coupled (or hybrid) system designs with oversized solar arrays, the ability to charge the coupled battery with grid energy, and larger batteries lead to the greatest capacity credit benefits of hybridization. We do not see evidence that the hybrid capacity credit exceeds the sum of the separate battery and solar capacity credits.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Tula: Optimizing Time, Cost, and Generalization in Distributed Large-Batch Training

Distributed training increases the number of batches processed per iteration either by scaling-out (adding more nodes) or scaling-up (increasing the batch-size). However, the largest configuration does not necessarily yield the best performance. Horizontal scaling introduces additional communication overhead, while vertical scaling is constrained by computation cost and device memory limits. Thus, simply increasing the batch-size leads to diminishing returns: training time and cost decrease initially but eventually plateaus, creating a knee-point in the time/cost vs. batch-size pareto curve. The optimal batch-size therefore depends on the underlying model, data and available compute resources. Large batches also suffer from worse model quality due to the well-known “generalization gap”. In this paper, we present Tula, an online service that automatically optimizes time, cost, and convergence quality for large-batch training of convolutional models. It combines parallel-systems modeling with statistical performance prediction to identify the optimal batchsize. Tula predicts training time and cost within 7.5−14% error across multiple models, and achieves up to 20× overall speedup and improves test accuracy by ≈9% on average over standard large-batch training on various vision tasks, thus successfully mitigating the generalization gap and accelerating training at the same time.

Tyagi, Sahil [ORNL] (ORCID:0009000783144745)↗