Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “learning rate”

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 1,603 records · Page 89
All sourcesNASA NTRSDOE OSTI
DOE OSTI2022

Ambient-temperature liquid jet targets for high-repetition-rate HED discovery science

High-power lasers can generate energetic particle beams and astrophysically relevant pressure and temperature states in the high-energy-density (HED) regime. Recently-commissioned high-repetition-rate (HRR) laser drivers are capable of producing these conditions at rates exceeding 1 Hz. However, experimental output from these systems is often limited by the difficulty of designing targets that match these repetition rates. To overcome this challenge, we have developed tungsten microfluidic nozzles, which produce a continuously replenishing jet that operates at flow speeds of approximately 10 m/s and can sustain shot frequencies up to 1 kHz. The ambient-temperature planar liquid jets produced by these nozzles can have thicknesses ranging from hundreds of nanometers to tens of micrometers. In this work, we illustrate the operational principle of the microfluidic nozzle and describe its implementation in a vacuum environment. Further, we provide evidence of successful laser-driven ion acceleration using this target and discuss the prospect of optimizing the ion acceleration performance through an in situ jet thickness scan. Future applications for the jet throughout HED science include shock compression and studies of strongly heated nonequilibrium plasmas. When fielded in concert with HRR-compatible laser, diagnostic, and active feedback technology, this target will facilitate advanced automated studies in HRR HED science, including machine learning-based optimization and high-dimensional statistical analysis.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2025

Converting sWeights to probabilities with density ratios

The use of machine learning approaches continues to have many benefits in experimental nuclear and particle physics. One common issue is generating training data which is sufficiently realistic to give reliable results. Here we advocate using real experimental data as the source of training data and demonstrate how one might subtract background contributions through the use of probabilistic weights which can be readily applied to training data. The sPlot formalism is a common tool used to isolate distributions from different sources. However, the negative sWeights produced by the sPlot technique can cause training problems and poor predictive power. This article demonstrates how density ratio estimation can be applied to convert sWeights to event probabilities, which we call drWeights. The drWeights can then be applied to produce the distributions of interest and are consistent with direct use of the sWeights. This article will also show how decision trees are particularly well suited to convert sWeights, with the benefit of fast prediction rates and adaptability to aspects of experimental data such as the data sample size and proportions of different event sources. We also show that a density ratio product approach in which the initial drWeights are reweighted by an additional converter gives substantially better results.

Glazier, Derek I. [Univ. of Glasgow, Scotland (Uni↗
NASA NTRS2005

Airdata Measurement and Calibration

This Section provides a brief introduction to airdata measurement and calibration. Readers will learn about typical test objectives, quantities to measure, and flight maneuvers and operations for calibration. The Section informs readers about tower-flyby, trailing cone, pacer, radar-tracking, and dynamic airdata calibration maneuvers. Readers will also begin to understand how some data analysis considerations and special airdata cases, including high-angle-of-attack flight, high-speed flight, and nonobtrusive sensors are handled. This section is not intended to be all inclusive; readers should review AGARDograph 300, Volume 1, "Calibration of Airdata Systems and Flow Direction Sensors" for more detailed information. [11-1] References 11-2, 11-3, and 11-4 also supply pertinent information to better understand airdata measurement and calibration and related terminology. Airdata are vital to successfully complete an aircraft's mission and are derived from the air surrounding the aircraft. These airdata encompass indicated and true airspeed, pressure altitude, ambient air temperature, angles of attack and sideslip, Mach number, and rate of climb. Typically, pitot and static pressures are sensed and converted (by mechanical means in the instruments themselves) into indications on the altimeter, vertical speed indicator, airspeed indicator, and Machmeter. Similarly, measured local flow angles establish angles of attack and sideslip, and the outside air temperature is measured and indicated in the cockpit. (Instruments that can perform the conversion, such as airspeed indicators, altimeters, and Machmeters, do not correct for errors in the input values.) These measured parameters are commonly input to the airdata computer which, using appropriate algorithms and correction factors (or calibrations, as discussed later), can provide other parameters, such as true airspeed, required by the aircraft's avionics or flight control system. The presence of the aircraft in the airstream causes input errors to the measuring instruments - the aircraft disturbs the air that it flies through, thereby also disturbing the airdata measurements. Figure 11-1 shows the airflow around an airplane wing. The air above the wing has lower pressure than the ambient air, while the pressure below the wing is higher than the ambient air. Compressibility and shock waves also disturb the air and affect the measurements. Compressibility effects become important above approximately Mach number 0.3. As a result the static pressure around an airplane varies considerably with location. Local flow angles also differ from the free-stream flow direction. In straight-and-level flight the airflow rises to the wing leading edge and falls below the trailing edge, causing errors in flow direction measurements. To some extent these errors can be studied in wind tunnels, but wind-tunnel measurements cannot replace in-flight measurements.

Haering, Edward A., Jr.↗
NASA NTRS2022

NASA Physics of Failure (PoF) for Reliability

An item’s reliability or longevity is dependent not only on its design but also on how it is used, manufactured, tested, and the stresses it has or will experience. Stresses include operational and environmental exposures to thermal, voltage, current, age/exposure, mechanical, and radiation mechanisms. Therefore, in reliability analysis, it is important to consider the contributions of all of these factors when predicting the failure rates of components. Historically, there has been a reliance on handbook data (e.g., MIL-HDBK-217), but experience has shown that these values and distributions are not representative of actual performance (1,2). Therefore, to make more credible reliability and risk assessments for its missions, NASA must transition to estimating likelihoods of failure based on an item’s reliability/longevity factors (or the physical susceptibilities and strengths impacting the design’s performance) has or will experience, whenever possible. To facilitate this transition a “Handbook on Methodology for Physics of Failure Based Reliability Assessments” has been developed by NASA to assist in applying physics experiences or experiment physics for empirical analysis and conceptualized physics exposures or theoretical physics for deterministic analysis, to develop and aggregate realistic likelihoods of failure leading to more credible forecasts of item performance and longevity. In addition, since it is NASA’s intention that this document continues to evolve based on community lessons learned and the introduction of new assessment methodologies, NASA is encouraging and appreciates the contributions of current and future authors to maintain and enhance this handbook and its supporting case studies.

Physics of Failure↗
NASA NTRS2024

An On-Board Off-Board Framework for Online Replanning: Applied to UAVs in Urban Environments

Autonomous systems are being used in a multitude of areas at an increasing rate and require a high level of adaptivity and intelligence to operate safely, especially under faulty conditions. This paper introduces a novel genetic algorithm tailored for UAV trajectory replanning, with an improved execution time via search space reduction based on the operating conditions of the UAV and its remaining mission. A unique characteristic of the replanning agent is its fast-start and adaptive properties, pre-seeding candidates with partial solutions and dynamically tuning elitism, crossover, and mutation rates in correspondence to the average fitness and diversity of the population. A population restart mechanism and early stopping mechanism are evaluated as well to assess their effect on solution quality and runtime. Previous work on genetic algorithms for UAV replanning were conducted with short trajectories in a small state space. Our UAV operates in a 56,000 square meter simulated urban environment, with static obstacles and a total of 53 possible waypoints. The agent increases the safety and reliability of UAV autonomy when operating under faulty conditions and when replanning is required.

Machine Learning↗
DOE OSTI2023

Scalable in situ non-destructive evaluation of additively manufactured components using process monitoring, sensor fusion, and machine learning

Laser Powder Bed Fusion (L-PBF) Additive Manufacturing (AM) is among the metal 3D printing technologies most broadly adopted by the manufacturing industry. However, the current industry qualification paradigm for critical-application L-PBF parts relies heavily on expensive non-destructive inspection techniques, which significantly limits the use-cases of L-PBF. In situ monitoring of the process promises a less expensive alternative to ex situ testing, but existing sensor technologies and data analysis techniques struggle to detect sub-surface flaws (e.g., porosity and cracking) on production-scale L-PBF printers. In this work, an in situ NDE (INDE) system was engineered to detect subsurface flaws detected in X-Ray Computed Tomography (XCT) directly from process monitoring data. A multilayer, multimodal data input allowed the INDE system to detect numerous subsurface flaws in the size range of 200–1000µm using a novel human-in-the-loop annotation procedure. Furthermore, a framework was established for generating probability-of-detection (POD) and probability-of-false-alarm (PFA) curves compliant with NDE standards by systematically comparing instances of detected subsurface flaws to post-build XCT data. Here, we also introduce for the first time in the AM in situ sensing literature the a 90/95 – the flaw size corresponding to a 90% detection rate on the lower 95% confidence interval of the POD curve. The INDE system successfully demonstrated POD capabilities commensurate with traditional NDE methods. Traditional ML performance metrics were also shown to be inadequate for assessing the ability of the INDE system’s flaw detection performance. It is the hope of the authors that future studies will adopt the POD and PFA approach outlined here to provide better insight into the utility of process monitoring for AM.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Uncertainty Quantification Framework for Predicting Material Response with Large Number of Parameters: Application to Creep Prediction in Ferritic-Martensitic Steels Using Combined Crystal Plasticity and Grain Boundary Models

This paper presents an uncertainty quantification (UQ) framework for the physics-based model prediction of material response with a large number of parameters. The application problem presented in this work is that of predicting creep in Grade 91 steel at 600°C. The material response is defined with a physically based microstructural model with constitutive equations emulating several observed phenomena in Grade 91 and embodied into an explicit geometry mesoscale finite element model for prior austenite grains and grain boundaries. Creep within the grains and in grain boundaries are represented by crystal plasticity for dislocation motion and a physics-based model for cavity growth and nucleation, respectively. The creep behavior of this material is influenced by several parameters, some of which have a wide range of variation based on experimental data. UQ combined with microstructural modeling can discover the core microstructural causes of experimental variability, leading to improved materials with lower variability in critical long-term material properties. In this study, we investigate the model's uncertainty to identify material properties that may be modified during production to increase creep life and analyze different components of the crystal plasticity model for improvements. For this purpose, a quantity of interest is defined as time to minimum creep rate, which correlates well to the creep failure of the material. A deep neural network model was trained and validated to be used as a surrogate for the finite element model. Then, a variance-based sensitivity analysis is performed on the surrogate model to find the Sobol indices of the input parameters in respect to the output quantity of interest. The Sobol indices are used to reduce the dimensionality of the model. Generalized polynomial chaos expansion is used on the reduced basis models to propagate the uncertainty from the input parameters to the quantity of interest using the deep neural network surrogate model. These results are benchmarked against uncertainty propagation using Monte Carlo simulations. In conclusion, the UQ performed through the reduced basis model captures almost all the uncertainty in the model with significantly fewer simulations, making it possible to perform the UQ directly via simulations with the finite element model rather than surrogate machine-learned models.

36 MATERIALS SCIENCE↗
DOE OSTI2022

OPALCO - Decatur Island Solar and Energy Storage Project: An Assessment of Battery Technical Performance

Orcas Power & Light Cooperative (OPALCO) is a member-owned, nonprofit cooperative utility that provides energy services to approximately 11,200 customers across 20 islands in San Juan County, Washington. OPALCO’s mostly hydroelectric power is generated by Bonneville Power Administration and delivered to the islands by submarine cables. In 2016, as part of the second round of funding from the Washington state Clean Energy Fund, OPALCO received a $1 million matching grant to support a project that deployed a 504-kW LG community photovoltaic (PV) system in combination with a 1 MW/2 MWh lithium-iron-phosphate battery energy storage system (BESS) on Decatur Island, Washington. The Decatur Island Substation is essential to ensuring reliable energy for the residents of the San Juan Islands as it is the point of interconnection with the mainland transmission system. The BESS, in combination with the community solar array, will deliver an innovative method to both defer the costly upgrade of the transmission system and allow for other high-value applications intended to benefit the utility and its customers. In 2018, Pacific Northwest National Laboratory (PNNL) completed a preliminary economic assessment for several identified use cases in collaboration with OPALCO. Between August 2021 and May 2022, extensive testing was conducted, and the results were used to assess the technical performance of the BESS subjected to actual field operations. Both reference performance and use case tests were performed: (a) Reference performance tests assess the general technical capabilities of the BESS, such as energy capacity, round-trip efficiency (RTE), ramp rate, and signal tracking capability. These are the first tests performed (baseline) and are repeated after use case tests (post cycle). A standardized U.S. Department of Energy (DOE) energy storage performance protocol was used to characterize the BESS, including representative duty cycle profiles, test procedure guidance, and calculation guidance for determining key characteristics. (b) Use case tests examine the performance of the BESS for specific use cases using duty cycles developed by PNNL in collaboration with OPALCO. Four use cases were selected for testing: 1) demand charge reduction, 2) load shaping, 3) outage mitigation, and 4) transmission deferral. The use case duty cycles were developed based on utility and site-specific characteristics in addition to the technical characteristics and physical capabilities of the BESS. Use case tests were performed between the baseline and post cycle tests. This report describes the BESS and its components, presents testing and performance analysis results, and shares key insights and lessons learned from this project. Outcomes of the tests and analyses will help OPALCO understand the performance of the Decatur Island BESS in its current state and design appropriate operational strategies for this and other BESSs over the long term.

14 SOLAR ENERGY↗
NASA NTRS2016

System Level Aerothermal Testing for the Adaptive Deployable Entry and Placement Technology (ADEPT)

The Adaptive Deployable Entry and Placement Technology (ADEPT), a mechanically deployable entry vehicle technology, has been under development at NASA since 2011. As part of the technical maturation of ADEPT, designs capable of delivering small payloads (10 kg) are being considered to rapidly mature sub 1 m deployed diameter designs. The unique capability of ADEPT for small payloads comes from its ability to stow within a slender volume and deploy to achieve a mass efficient drag surface with a high heat rate capability. The low ballistic coefficient results in entry heating and mechanical loads that can be met by a revolutionary three-dimensionally woven carbon fabric supported by a deployable skeleton structure. This carbon fabric has test proven capability as both primary structure and payload thermal protection system. In order to rapidly advance ADEPTs technical maturation, the project is developing test methods that enable thermostructural design requirement verification of ADEPT designs at the system level using ground test facilities. Results from these tests are also relevant to larger class missions and help us define areas of focused component level testing in order to mature material and thermal response design codes. The ability to ground test sub 1 m diameter ADEPT configurations at or near full-scale provides significant value to the rapid maturation of this class of deployable entry vehicles. This paper will summarize arc jet test results, highlight design challenges, provide a summary of lessons learned and discuss future test approaches based upon this methodology.

Hypersonic↗
DOE OSTI2025

Net Present Value Optimization of a Natural Gas Combined Cycle Plant with CO 2 Capture using a Water-Lean Solvent Considering Transient Electricity Price for Multiple Regions

Global CO 2 emissions are increasing at about a 1.5% rate per year. Fossil fuel-based plants are one of the main contributors to this rise. In the power generation industry, fossil fuel plants are dominant, and many plants are under development. In this study, a natural gas combined cycle (NGCC) power plant with postcombustion capture using a leading water-lean solvent is considered. For optimal design and operating schedule, large-scale dynamic optimization is undertaken for net present value (NPV) optimization. The first principle dynamic model of NGCC is developed, including a model of the highly efficient H-class gas turbines. For computational tractability of the dynamic optimization problem, a reduced-order model is developed by using the Hankel singular value decomposition. A waterlean solvent, N-(2-ethoxyethyl)-3-morpholinopropan-1-amine, is used for carbon capture. A model of the capture system is developed in Aspen Plus, which is used to develop a reduced-order model by using ALAMO, a machine learning software. In addition, a reduced model of the CO 2 compression system with a dehydration unit is also considered. The integrated system is used for NPV optimization by using the Python-based PYOMO platform. The PCC process is analyzed for three configurations-conventional packed bed, rotating packed bed (RPB), and a combination of RPB and direct contact cooler. The NPV optimization is performed for 14 regional markets by considering year-long clustered and continuous locational marginal price data with a 1 h interval. Optimization results show that the PCC can achieve 90% CO 2 capture with a positive NPV for six regions. Sensitivity studies conducted by using the PCC configurations indicate that the process is economically feasible for 9 regions out of 14 regional electricity markets with NPV values in the range of 33−540 $MM.

cabon capture↗
DOE OSTI2022

Machine-learning-guided descriptor selection for predicting corrosion resistance in multi-principal element alloys

More than $270 billion is spent on combatting corrosion annually in the USA alone. As such, we present a machine-learning (ML) approach to down select corrosion-resistant alloys. Our focus is on a non-traditional class of alloys called multi-principal element alloys (MPEAs). Given the vast search space due to the variety of compositions and descriptors to be considered, and based upon existing corrosion data for MPEAs, we demonstrate descriptor optimization to predict corrosion resistance of any given MPEA. Our ML model with descriptor optimization predicts the corrosion resistance of a given MPEA in the presence of an aqueous environment by down selecting two environmental descriptors (pH of the medium and halide concentration), one chemical composition descriptor (atomic % of element with minimum reduction potential), and two atomic descriptors (difference in lattice constant (Δa) and average reduction potential). Our findings show that, while it is possible to down select corrosion-resistant MPEAs by using ML from a large search space, a larger dataset and higher quality data are needed to accurately predict the corrosion rate of MPEAs. This study shows both the promise and the perils of ML when applied to a complex chemical phenomenon like corrosion of alloys.

36 MATERIALS SCIENCE↗
DOE OSTI2026

Crystal generation using the fully differentiable pipeline and latent space optimization

We present a materials generation framework that couples a symmetry-conditioned variational autoencoder with a differentiable SO(3) power spectrum objective to steer candidates toward a specified local environment under the crystallographic constraints. In particular, we implement a fully differentiable pipeline that performs batch-wise optimization on both direct and latent crystallographic representations. Using the GPU acceleration, the implementation achieves about fivefold speed compared to our previous CPU workflow, while yielding comparable outcomes. In addition, we introduce the optimization strategy that alternatively performs optimization on the direct and latent crystal representations. This dual-level relaxation approach can effectively escape local minima defined by different objective gradients, thus increasing the success rate of generating complex structures satisfying the target local environments. This framework can be extended to systems consisting of multi-components and multi-environments, providing a scalable route to generate material structures with the target local environment.

conditional VAE↗
DOE OSTI2023

Wave Detection and Tracking Within a Rotating Detonation Engine Through Object Detection

As the operational time window of experimental rotating detonation engines (RDEs) is expanded and the technology matures toward integration within gas turbines, monitoring techniques must evolve to offer computationally efficient and highly time-resolved diagnostics. In this study, computer vision object detection methodology that seeks to reduce data processing time and calculate wave velocity within drastically reduced time intervals as compared to traditional high-frame-rate RDE images analysis techniques is proposed. The adapted you-only-look-once object detection network is trained to detect individual detonation waves within single down-axis RDE images. The wave location and rotational direction detected within a frame are tracked through a series of high-speed images to calculate the frame-to-frame wave velocity with the time-step resolution of $\mathrm{20 μs}$ across a series of frames. The analysis of the annotation box size and image linearization effects is presented, demonstrating the lowest frame-to-frame velocity total uncertainty of $\mathrm{±3.8\%}$ and the highest classification speed of 9.5 frames per second using linearized images. Linearized images “unwrap” the RDE annulus pixel region to a reduced image size. Here, this new method offers great reductions in data processing times and unsteady detonation behavior insight at intervals more comparable to the timescales of detonation wave interactions via the application of machine learning to experimental RDE data.

33 ADVANCED PROPULSION SYSTEMS↗
NASA NTRS1986

At least some errors are randomly generated (Freud was wrong)

An experiment was carried out to expose something about human error generating mechanisms. In the context of the experiment, an error was made when a subject pressed the wrong key on a computer keyboard or pressed no key at all in the time allotted. These might be considered, respectively, errors of substitution and errors of omission. Each of seven subjects saw a sequence of three digital numbers, made an easily learned binary judgement about each, and was to press the appropriate one of two keys. Each session consisted of 1,000 presentations of randomly permuted, fixed numbers broken into 10 blocks of 100. One of two keys should have been pressed within one second of the onset of each stimulus. These data were subjected to statistical analyses in order to probe the nature of the error generating mechanisms. Goodness of fit tests for a Poisson distribution for the number of errors per 50 trial interval and for an exponential distribution of the length of the intervals between errors were carried out. There is evidence for an endogenous mechanism that may best be described as a random error generator. Furthermore, an item analysis of the number of errors produced per stimulus suggests the existence of a second mechanism operating on task driven factors producing exogenous errors. Some errors, at least, are the result of constant probability generating mechanisms with error rate idiosyncratically determined for each subject.

Sellen, A. J.↗
DOE OSTI2022

Autonomous Wireless Technology Detection in Seamless IoT Applications

The ever-increasing use of Internet of Things (IoT) devices results in the implementation of multiple wireless technologies that would not only cater their data rate requirements but also support various applications. To optimize the energy efficiency and security of the wireless transmission, it is imperative to identify the wireless technologies in various IoT implementations. Many of the existing approaches are based on measuring only the receiving signal strength indicator (RSSI). However, such approaches may not work well because of transmit power control and complex channel variations among different wireless technologies. In this article, we propose an autonomous wireless detection scheme that considers multiple distinguishable physical (PHY)-layer settings for real-time identification of wireless technologies for real-time applications. Specifically, the proposed scheme relies on the PHY-layer measurements of the targeted spectrum. Transmission settings, such as bandwidth, carrier frequency, and RSSI are estimated from the raw in-phase and quadrature-phase (I/Q) measurements. In addition, a symbol-level extraction scheme is implemented to extract unique features of modulation settings. These aforementioned features are applied to a machine learning process to identify the received wireless technologies. Compared with raw I/Q measurements, the extracted features are much simplified and, thus, the machine learning classifier can be designed with a simple structure for fast processing on IoT nodes. Finally, the proposed schemes are primarily evaluated theoretically, followed by implementing them on a USRP software-defined radio (SDR)-based hardware testbed. The evaluation results demonstrate high accuracy in the real-time detection of different wireless technologies for seamless IoT applications.

42 ENGINEERING↗
NASA NTRS2011

Achieving the Proper Balance Between Crew and Public Safety

A paramount objective of all human-rated launch and reentry vehicle developers is to ensure that the risks to both the crew onboard and the public are minimized within reasonable cost, schedule, and technical constraints. Past experience has shown that proper attention to range safety requirements necessary to ensure public safety must be given early in the design phase to avoid additional operational complexities or threats to the safety of people onboard, and the design engineers must give these requirements the same consideration as crew safety requirements. For human spaceflight, the primary purpose and operational concept for any flight safety system is to protect the public while maximizing the likelihood of crew survival. This paper will outline the policy considerations, technical issues, and operational impacts regarding launch and reentry vehicle failure scenarios where crew and public safety are intertwined and thus addressed optimally in an integrated manner. An overview of existing range and crew safety policy requirements will be presented. Application of these requirements and lessons learned from both the Space Shuttle and Constellation Programs will also be discussed. Using these past programs as examples, the paper will detail operational, design, and analysis approaches to mitigate and balance the risks to people onboard and in the public. Manned vehicle perspectives from the Federal Aviation Administration (FAA) and Air Force organizations that oversee public safety will be summarized as well. Finally, the paper will emphasize the need to factor policy, operational, and analysis considerations into the early design trades of new vehicles to help ensure that both crew and public safety are maximized to the greatest extent possible.

Gowan, John↗
NASA NTRS2008

The Ares Project: Building an Exploration Culture from the Inside Up

NASA is building its first new human-rated space exploration vehicles in nearly 40 years. This marks an important operational and cultural change from the Space Shuttle. In the wake of the Columbia disaster, the agency and the nation realized that NASA's goals and culture needed to change. The Ares Project, which is building the launch vehicles that will power human beings to the Moon, Mars, and beyond, is taking a page from the Saturn playbook by having NASA lead both the overall integration and the development of the Ares I upper stage. Ares is also creating a new culture of cooperation, openness, and informed risk taking as we set our sights on other worlds. Ares has established a team environment where issues can be discussed, information is shared, fun and teamwork are encouraged, and constructive conflict and accountability are expected. Following a "One NASA" philosophy, Ares is taking steps to strengthen cooperation among space centers, contractor partners, engineering and scientific communities, and headquarters personnel. As we learn lessons from things that went wrong with the Space Shuttle, we are also borrowing best practices from what has gone right with that program and others. All of these cultural elements will be necessary as we take the next steps beyond Earth orbit.

Cook, Stephan A.↗
DOE OSTI2025

FedOSAA: Improving Federated Learning with One-Step Anderson Acceleration

Federated learning (FL) is a distributed machine learning approach that enables multiple local clients and a central server to collaboratively train a model while keeping the data on their own devices. First-order methods, particularly those incorporating variance reduction techniques, are the most widely used FL algorithms due to their simple implementation and stable performance. However, these methods tend to be slow and require a large number of communication rounds to reach the global minimizer. We propose FedOSAA, a novel approach that preserves the simplicity of first-order methods while achieving the rapid convergence typically associated with second-order methods. Our approach applies one Anderson acceleration (AA) step following classical local updates based on first-order methods with variance reduction, such as FedSVRG and SCAFFOLD, during local training. This AA step is able to leverage curvature information from the history points and gives a new update that approximates the Newton-GMRES direction, thereby significantly improving the convergence. We establish a local linear convergence rate to the global minimizer of FedOSAA for smooth and strongly convex loss functions. Numerical comparisons show that FedOSAA substantially improves the communication and computation efficiency of the original first-order methods, achieving performance comparable to second-order methods like GIANT.

Feng, Xue [University of California, Davis]↗