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 937 records · Page 52
All sourcesNASA NTRSDOE OSTI
NASA NTRS2016

High-Performance Reaction Wheel Optimization for Fine-Pointing Space Platforms: Minimizing Induced Vibration Effects on Jitter Performance plus Lessons Learned from Hubble Space Telescope for Current and Future Spacecraft Applications

The Hubble Space Telescope (HST) applies large-diameter optics (2.5-m primary mirror) for diffraction-limited resolution spanning an extended wavelength range (approx. 100-2500 nm). Its Pointing Control System (PCS) Reaction Wheel Assemblies (RWAs), in the Support Systems Module (SSM), acquired an unprecedented set of high-sensitivity Induced Vibration (IV) data for 5 flight-certified RWAs: dwelling at set rotation rates. Focused on 4 key ratios, force and moment harmonic values (in 3 local principal directions) are extracted in the RWA operating range (0-3000 RPM). The IV test data, obtained under ambient lab conditions, are investigated in detail, evaluated, compiled, and curve-fitted; variational trends, core causes, and unforeseen anomalies are addressed. In aggregate, these values constitute a statistically-valid basis to quantify ground test-to-test variations and facilitate extrapolations to on-orbit conditions. Accumulated knowledge of bearing-rotor vibrational sources, corresponding harmonic contributions, and salient elements of IV key variability factors are discussed. An evolved methodology is presented for absolute assessments and relative comparisons of macro-level IV signal magnitude due to micro-level construction-assembly geometric details/imperfections stemming from both electrical drive and primary bearing design parameters. Based upon studies of same-size/similar-design momentum wheels' IV changes, upper estimates due to transitions from ground tests to orbital conditions are derived. Recommended HST RWA choices are discussed relative to system optimization/tradeoffs of Line-Of-Sight (LOS) vector-pointing focal-plane error driven by higher IV transmissibilities through low-damped structural dynamics that stimulate optical elements. Unique analytical disturbance results for orbital HST accelerations are described applicable to microgravity efforts. Conclusions, lessons learned, historical context/insights, and perspectives on future applications are given; these previously unpublished data and findings represents a valuable resource for fine-pointing spacecraft or space-based platforms using RWAs, Control Moment Gyros (CMGs), Momentum Wheels, or other ball-bearing-based rotational units.

Hasha, Martin D.↗
DOE OSTI2023

Machine-Learning Assisted Identification of Battery Life Models

Predictive battery life models are commonly utilized to extrapolate degradation trends observed during accelerated aging tests for simulation of degradation in real-world applications. Thus, fitting accelerated aging data as accurately as possible and with low uncertainty is crucial for making believable projections of battery lifetime, but it is challenging to identify algebraic expressions that accurately fit multivariate degradation trends. A review of models published in literature reveal some common expressions for fitting calendar aging data, which is only dependent on temperature and state-of-charge, but no consistency across many models for fitting cycle aging data, indicating the need for a statistically rigorous data driven approach for developing empirical models. This talk will describe a machine-learning assisted method for identification of predictive battery life models utilizing bilevel optimization and symbolic regression. Bilevel optimization with cross-validation is used to statistically determine cell- and stress-dependent model parameters, while symbolic regression identifies both linear and multiplicative candidate expressions to predict stress-dependent degradation rates by selecting low-order subsets of features from a generated feature library. Because model expressions are identified empirically, it is crucial to ensure resulting models behave according to physical expectations, so the stability of models for interpolation or extrapolation is interrogated qualitatively through simulation and quantitatively through cross-validation and uncertainty quantification via bootstrap resampling. This model identification approach substantially improves upon models identified purely using expert judgement in terms of both accuracy and uncertainty. Model simulation and validation is then conducted by deriving a state-equation form of the predictive model, enabling simulation of battery aging under dynamic stresses. This enables validation of the predictive battery model on lab-based tests with varying conditions or on drive-cycle or application-cycle testing protocols. Parameter uncertainty can be carried forward into model simulation, giving lifetime estimates and confidence windows for cell- or system-level lifetime. The financial impact of battery model uncertainty can be estimated by incorporating uncertainty into a technoeconomic model.

battery↗
DOE OSTI2025

Harnessing Machine Learning to Predict MoS 2 Solid Lubricant Performance

Physical vapor deposited (PVD) molybdenum disulfide (MoS 2 ) solid lubricant coatings are an exemplar material system for machine learning methods due to small changes in process variables often causing large variations in microstructure and mechanical/tribological properties. Here, in this work, a gradient boosted regression tree machine learning method is applied to an existing experimental data set containing process, microstructure, and property information to create deeper insights into the process-structure–property relationships for molybdenum disulfide (MoS 2 ) solid lubricant coatings. The optimized and cross-validated models show good predictive capabilities for density, reduced modulus, hardness, wear rate, and initial coefficients of friction. The contribution of individual deposition variables (i.e., argon pressure, deposition power, target conditioning) on coating properties is highlighted through feature importance. The process-property relationships established herein show linear and non-linear relationships and highlight the influence of uncontrolled deposition variables (i.e., target conditioning) on the tribological performance.

MoS2↗
DOE OSTI2025

Performance evaluations of signed and unsigned noisy approximate quantum Fourier arithmetic

The Quantum Fourier Transform (QFT) grants competitive advantages, especially in resource usage and circuit approximation, for performing arithmetic operations on quantum computers, and offers a potential route toward a numerical quantum-computational paradigm. In this paper, we utilize efficient techniques to implement QFT-based integer addition and multiplications. These operations are fundamental to various quantum applications including Shor’s algorithm, weighted-sum optimization problems in data processing and machine learning, and quantum algorithms requiring inner products. We carry out performance evaluations of these implementations based on IBM’s superconducting-qubit architecture using different compatible noise models. We isolate the sensitivity of the component quantum circuits on both one-/two-qubit gate error rates, and the number of the arithmetic operands’ superposed integer states. We analyze performance and identify the most effective approximation depths for unsigned quantum addition and quantum multiplication within the given context. We then perform a similar analysis of signed addition and compare to the unsigned results. We observe significant dependency of the optimal approximation depth on the degree of machine noise and the number of superposed states in certain performance regimes. Finally, we elaborate on the algorithmic challenges—relevant to signed, unsigned, modular and non-modular versions—that could also be applied to current implementations of QFT-based subtraction, division, exponentiation, and their potential tensor extensions. Here, we analyze the performance trends in our results and speculate on possible future developments within this computational paradigm.

Computational models↗
DOE OSTI2026

Concerted Electron-Ion Transport by Polyacrylonitrile Elucidated with Reactive Deep Learning Potentials

Charge transport in polymers, such as polyacrylonitrile (PAN), is crucial for electronics and energy storage. For instance, PAN can transport cations e.g., Li + , by facilitating dynamic cation-nitrile coordination in batteries. However, little is known regarding the underlying role of complex reactive polymer configurations. Herein, we develop a deep-learning potential, trained on ab initio energies and forces of nonequilibrium reactive PAN configurations, to unravel the kinetics of PAN cyclization initiated by a nucleophile (OH – dissociated from LiOH) attacking the terminal nitrile carbon. We find, based on the reaction free-energetics, rates, and charge analysis, that the nucleophile attack producing the first ring is the rate-limiting step, which subsequently triggers Li + -coupled electron transfer along the PAN backbone, causing ∼10 4 times faster sequential ring-formation of the remaining nitriles. PAN’s extended configurations, where dipolar and H-bonding interactions are minimal, enable such rapid kinetics. By validating our computational findings with IR and NMR experiments, we establish a pathway for designing reactive polymers with enhanced charge transport for energy applications.

Chahal-Crockett, Rajni [Oak Ridge National Laborat↗
DOE OSTI2024

Exploring the effects of molecular beam epitaxy growth characteristics on the temperature performance of state-of-the-art terahertz quantum cascade lasers

This study conducts a comparative analysis, using non-equilibrium Green’s functions (NEGF), of two state-of-the-art two-well (TW) Terahertz Quantum Cascade Lasers (THz QCLs) supporting clean 3-level systems. The devices have nearly identical parameters and the NEGF calculations with an abrupt-interface roughness height of 0.12 nm predict a maximum operating temperature (T max ) of ~ 250 K for both devices. However, experimentally, one device reaches a T max of ~ 250 K and the other a T max of only ~ 134 K. Both devices were fabricated and measured under identical conditions in the same laboratory, with high quality processes as verified by reference devices. The main difference between the two devices is that they were grown in different MBE reactors. Our NEGF-based analysis considered all parameters related to MBE growth, including the maximum estimated variation in aluminum content, growth rate, doping density, background doping, and abrupt-interface roughness height. From our NEGF calculations it is evident that the sole parameter to which a drastic drop in T max could be attributed is the abrupt-interface roughness height. We can also learn from the simulations that both devices exhibit high-quality interfaces, with one having an abrupt-interface roughness height of approximately an atomic layer and the other approximately a monolayer. However, these small differences in interface sharpness are the cause of the large performance discrepancy. This underscores the sensitivity of device performance to interface roughness and emphasizes its strategic role in achieving higher operating temperatures for THz QCLs. We suggest Atom Probe Tomography (APT) as a path to analyze and measure the (graded)-interfaces roughness (IFR) parameters for THz QCLs, and subsequently as a design tool for higher performance THz QCLs, as was done for mid-IR QCLs. Our study not only addresses challenges faced by other groups in reproducing the record T max of ~ 250 K and ~ 261 K but also proposes a systematic pathway for further improving the temperature performance of THz QCLs beyond the state-of-the-art.

77 NANOSCIENCE AND NANOTECHNOLOGY↗
DOE OSTI2021

Quenching, bursting, and galaxy shapes: colour transformation as a function of morphology

ABSTRACT Different mechanisms for quenching star formation in galaxies are commonly invoked in the literature, but the relative impact of each one at different cosmic epochs is still unknown. In particular, the relation between these processes and morphological transformation remains poorly understood. In this work, we measure the effectiveness of changes in star formation rates by analysing a new parameter, the Star Formation Acceleration (SFA), as a function of galaxy morphology. This methodology is capable of identifying both bursting and quenching episodes that occurred in the preceding 300 Myr. We use morphological classification catalogues based on Deep Learning techniques. Our final sample has ∼14 200 spirals and ∼2500 ellipticals. We find that elliptical galaxies in the transition region have median shorter quenching time-scales (τ < 1 Gyr) than spirals (τ ≥ 1 Gyr). This result conforms to the scenario in which major mergers and other violent processes play a fundamental role in galaxy evolution for most ellipticals, not only quenching star formation more rapidly but also playing a role in morphological transformation. We also find that ∼two-thirds of galaxies bursting in the green valley in our sample are massive spirals ($M_\star \ge 10^{11.0}\, \mathrm{M}_\odot$) with signs of disturbance. This is in accordance with the scenario where low-mass galaxies are losing their gas in an interaction with a massive galaxy: while the former is quenching, the last is being refueled and going through a burst, showing signs of recent interaction.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2023

Precursor recommendation for inorganic synthesis by machine learning materials similarity from scientific literature

Synthesis prediction is a key accelerator for the rapid design of advanced materials. However, determining synthesis variables such as the choice of precursor materials is challenging for inorganic materials because the sequence of reactions during heating is not well understood. In this work, we use a knowledge base of 29,900 solid-state synthesis recipes, text-mined from the scientific literature, to automatically learn which precursors to recommend for the synthesis of a novel target material. The data-driven approach learns chemical similarity of materials and refers the synthesis of a new target to precedent synthesis procedures of similar materials, mimicking human synthesis design. When proposing five precursor sets for each of 2654 unseen test target materials, the recommendation strategy achieves a success rate of at least 82%. Our approach captures decades of heuristic synthesis data in a mathematical form, making it accessible for use in recommendation engines and autonomous laboratories.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Neutron Tagging From Neutrino Interactions in DUNE-ND 2x2 Prototype

The Deep Underground Neutrino Experiment (DUNE) is a long-baseline neutrino oscillation experiment that aims to measure whether CP is violated in the leptonic sector (if violated, how much) and unambiguously determine the neutrino mass ordering. DUNE consists of near and far detectors that rely on liquid argon time projection chamber (LArTPC) technology to observe neutrino interactions. The near detector (ND) will be placed in Fermilab, near the neutrino source, while the far detector (FD) will be deployed in Sanford Lab, 1.5 km deep underground, which is 1300 km away from the source. LArTPCs provide excellent particle identification and calorimetry; however, detecting neutrons is challenging, as they do not leave direct ionization signals in LArTPCs. Neutrons can carry away up to 25% of the neutrino energy, introducing a significant uncertainty in DUNE measurements. The DUNE near detector (ND) features a novel modular LArTPC with pixelated charge readout, which enhances event recons truction. The modular design enables precise correlation between ionization signals and light signals in a high-rate environment, improving the identification of delayed energy depositions from neutrons in neutrino interactions. We introduce a neutron tagging technique using the 2x2 Demonstrator, a small-scale prototype of the DUNE ND LArTPC. The analysis utilizes Monte Carlo simulations and deep-learning techniques to identify neutron-induced energy depositions and reconstruct low-energy activity.

Kufatty, Georgette [Florida State U.]↗
NASA NTRS2001

Microarthroscopy System With Image Processing Technology Developed for Minimally Invasive Surgery

In a joint effort, NASA, Micro Medical Devices, and the Cleveland Clinic have developed a microarthroscopy system with digital image processing. This system consists of a disposable endoscope the size of a needle that is aimed at expanding the use of minimally invasive surgery on the knee, ankle, and other small joints. This device not only allows surgeons to make smaller incisions (by improving the clarity and brightness of images), but it gives them a better view of the injured area to make more accurate diagnoses. Because of its small size, the endoscope helps reduce physical trauma and speeds patient recovery. The faster recovery rate also makes the system cost effective for patients. The digital image processing software used with the device was originally developed by the NASA Glenn Research Center to conduct computer simulations of satellite positioning in space. It was later modified to reflect lessons learned in enhancing photographic images in support of the Center's microgravity program. Glenn's Photovoltaic Branch and Graphics and Visualization Lab (G-VIS) computer programmers and software developers enhanced and speed up graphic imaging for this application. Mary Vickerman at Glenn developed algorithms that enabled Micro Medical Devices to eliminate interference and improve the images.

Steele, Gynelle C.↗
NASA NTRS2017

NASA's Bio-Inspired Acoustic Absorber Concept

Transportation noise pollutes our worlds cities, suburbs, parks, and wilderness areas. NASAs fundamental research in aviation acoustics is helping to find innovative solutions to this multifaceted problem. NASA is learning from nature to develop the next generation of quiet aircraft.The number of road vehicles and airplanes has roughly tripled since the 1960s. Transportation noise is audible in nearly all the counties across the US. Noise can damage your hearing, raise your heart rate and blood pressure, disrupt your sleep, and make communication difficult. Noise pollution threatens wildlife when it prevents animals from hearing prey, predators, and mates. Noise regulations help drive industry to develop quieter aircraft. Noise standards for aircraft have been developed by the International Civil Aviation Organization and adopted by the US Federal Aviation Administration. The US National Park Service is working with the Federal Aviation Administration to try to balance the demand for access to the parks and wilderness areas with preservation of the natural soundscape. NASA is helping by conceptualizing quieter, more efficient aircraft of the future and performing the fundamental research to make these concepts a reality someday. Recently, NASA has developed synthetic structures that can absorb sound well over a wide frequency range, and particularly below 1000 Hz, and which mimic the acoustic performance of bundles of natural reeds. We are adapting these structures to control noise on aircraft, and spacecraft. This technology might be used in many other industrial or architectural applications where acoustic absorbers have tight constraints on weight and thickness, and may be exposed to high temperatures or liquids. Information about this technology is being made available through reports and presentations available through the NASA Technical Report Server, http:ntrs.nasa.gov. Organizations who would like to collaborate with NASA or commercialize NASAs technology are encouraged to contact the NASA Glenn Technology Transfer Office, https:technology.grc.nasa.gov. The NASA Glenn Office of Education https:www.nasa.govcentersglenneducationindex.html and the NASA Glenn Virtual Interchange for Nature-Inspired Exploration https:www.grc.nasa.govvine are also helping to make research like this accessible to the public and students of all ages.

Acoustics↗
NASA NTRS2016

Time Series Explorer: Bayesian Blocks with Generalized Profiles and in Higher Dimensions

The Time Series Explorer (TSE) is a project aimed at provided new and advanced time series analysis algorithms in two forms: a tool kit and an automated pipeline applying selected tools in machine learning settings. I will present a sketch of TSE with emphasis on time-domain modeling in general and recent improvements of the Bayesian Block (BB) algorithm in particular. This includes generalizing the shape of the elementary blocks from the current constant-rate model to general shapes, such as two-sided exponentials. Related topics will include extension of BB to higher dimensions and a novel way to detect and characterize short time-scale bursts in time-tagged event data. The Fermi Gamma Ray Space Telescope light curve for the Crab Nebula will be used as an example for all of the algorithms discussed. This work is in collaboration with Tom Loredo.

Bayesian Block (BB) algorithm↗
NASA NTRS2018

A 3D Citizen Science Video Game for NeMO-Net, the NASA Neural Multi-Modal Observation and Training Network for Global Coral Reef Assessment

NeMO-Net, the NASA neural multi-modal observation and training network for global coral reef assessment, is an open-source deep convolutional neural network aimed at accurately assessing the present and past dynamics of coral reef ecosystems through determination of percent living cover and morphology. We present here the active learning component of the project, which consists of an interactive video game prototype for tablet and mobile devices where players are able to intuitively label morphology classifications over mm-scale 3D coral reef imagery. Active learning applications present a novel methodology for engaging the public while efficiently providing large-scale training and test data for increasingly complex and data-intensive machine learning algorithms. NeMO-Net trains players on domain-specific knowledge through interactive tutorials and periodically checks players' input against pre-classified coral imagery to gauge their accuracy and utilize in-game mechanics to provide personalized classification training. Players can rate the classifications of other players, unlock rewards and join a global community as they explore and classify coral reefs and other shallow marine environments.

Citizen Science↗
DOE OSTI2025

Machine Learning for Predicting Multipactor Susceptibility in Planar RF Structures

Multipactor discharge is a persistent challenge in high-power microwave (HPM) and accelerator systems, where secondary electron avalanches can cause heating, vacuum degradation, and failure. This work presents the first supervised machine learning (ML) framework for multipactor prediction, trained on high-fidelity 3D Particle-in-Cell (PIC) simulation data in planar geometries. The model maps operational, geometric, and material-dependent secondary electron yield (SEY) parameters to the time-averaged electron growth rate, enabling rapid reconstruction of susceptibility charts. Among the models evaluated, tree-based ensemble methods such as Random Forest and Extra Trees demonstrate superior generalization to unseen materials compared to neural networks such as multilayer perceptron (MLP). Performance metrics, including Intersection over Union (IoU), Structural Similarity Index Measure (SSIM), and Pearson correlation, show close agreement with simulation benchmarks. Principal Component Analysis attributes generalization limits to material feature-space disjointedness.

43 PARTICLE ACCELERATORS↗
DOE OSTI2022

ARETE: Accurate Error Assessment via Machine Learning-Guided Dynamic-Timing Analysis

Nanometer circuits are increasingly prone to timing errors, escalating the need for fault injection frameworks to accurately evaluate their impact on applications. Here in this paper, we propose ARETE, a novel cross-layer, fault-injection framework that combines dynamic-binary instrumentation with machine learning-guided dynamic-timing analysis. ARETE enables accurate fault-injection into any application by estimating the location of the injecting errors via dynamic-timing analysis. To accelerate fault-injection, we develop a novel, data-aware, machine learning-based mechanism that dynamically pre-selects the error-prone instructions and limits the application of the costly dynamic-timing analysis only to them. To evaluate ARETE's accuracy, our fully automated toolflow is configured to support fault-injection based on detailed post-layout gate-level simulations as well as via existing workload-agnostic error models. Our results for various workloads, including an autonomous-driving library, show that the location and time of injected errors performed by ARETE, is 89.9% consistent with fault-injection based on full gate-level simulation. On average, ARETE executes 84.6x faster than gate-level simulation and at a cost of 3.4% loss in the program output quality estimation. When compared to the existing statistical fault-injection tools that are based on workload-agnostic error models, ARETE improves the accuracy of fault-injection rate and output quality estimation by 143.9% and 40.4% on average, respectively.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Airborne hyperspectral imaging of nitrogen deficiency on crop traits and yield of maize by machine learning and radiative transfer modeling

Nitrogen is an essential nutrient that directly affects plant photosynthesis, crop yield, and biomass production for bioenergy crops, but excessive application of nitrogen fertilizers can cause environmental degradation. To achieve sustainable nitrogen fertilizer management for precision agriculture, there is an urgent need for nondestructive and high spatial resolution monitoring of crop nitrogen and its allocation to photosynthetic proteins as that changes over time. Here, we used visible to shortwave infrared (400–2400 nm) airborne hyperspectral imaging with high spatial (0.5 m) and spectral (3–5 nm) resolutions to accurately estimate critical crop traits, i.e., nitrogen, chlorophyll, and photosynthetic capacity (CO 2 -saturated photosynthesis rate, V max,27 ), at leaf and canopy scales, and to assess nitrogen deficiency on crop yield. We conducted three airborne campaigns over a maize (Zea mays L.) field during the growing season of 2019. Physically based soil-canopy Radiative Transfer Modeling (RTM) and data-driven approaches i.e. Partial-Least Squares Regression (PLSR) were used to retrieve crop traits from hyperspectral reflectance, with ground truth of leaf nitrogen, chlorophyll, V max,27 , Leaf Area Index (LAI), and harvested grain yield. To improve computational efficiency of RTMs, Random Forest (RF) was used to mimic RTM simulations to generate machine learning surrogate models RTM-RF. The results show that prior knowledge of soil background and leaf angle distribution can significantly reduce the ill-posed RTM retrieval. RTM-RF achieved a high accuracy to predict leaf chlorophyll content (R 2 = 0.73) and LAI (R 2 = 0.75). Meanwhile, PLSR exhibited better accuracy to predict leaf chlorophyll content (R 2 = 0.79), nitrogen concentration (R 2 = 0.83), nitrogen content (R 2 = 0.77), and V max,27 (R 2 = 0.69) but required measured traits for model training. We also found that canopy structure signals can enhance the use of spectral data to predict nitrogen related photosynthetic traits, as combining RTM-RF LAI and PLSR leaf traits well predicted canopy-level traits (leaf traits × LAI) including canopy chlorophyll (R 2 = 0.80), nitrogen (R 2 = 0.85) and V max,27 (R 2 = 0.82). Compared to leaf traits, we further found that canopy-level photosynthetic traits, particularly canopy V max,27 , have higher correlation with maize grain yield. This study highlights the potential for synergistic use of process-based and data-driven approaches of hyperspectral imaging to quantify crop traits that facilitate precision agricultural management to secure food and bioenergy production.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2022

Short–Term Forecasting of Wind Gusts at Airports Across CONUS Using Machine Learning

Short–term forecasting of wind gusts, particularly those of higher intensity, is of great societal importance but is challenging due to the presence of multiple gust generation mechanisms. Wind gust observations from eight high–passenger–volume airports across the continental United States (CONUS) are summarized and used to develop predictive models of wind gust occurrence and magnitude. These short–term (same hour) forecast models are built using multiple logistic and linear regression, as well as artificial neural networks (ANNs) of varying complexity. A suite of 19 upper–air predictors drawn from the ERA5 reanalysis and an autoregressive (AR) term are used. Stepwise procedures instruct predictor selection, and resampling is used to quantify model stability. All models are developed separately for the warm (April–September) and cold (October–March) seasons. Results show that ANNs of 3–5 hidden layers (HLs) generally exhibit higher hit rates than logistic regression models and also improve skill with respect to wind gust magnitudes. However, deeper networks with more HLs increase false alarm rates in occurrence models and mean absolute error in magnitude models due to model overfitting. For model skill, inclusion of the AR term is critical while the majority of the remaining skill derives from wind speeds and lapse rates. A predictive ceiling is also clearly demonstrated, particularly for the strong and damaging gust magnitudes, which appears to be partially due to ERA5 predictor characteristics and the presence of mixed wind climates.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2026

Quantum-Inspired Bayesian Sampling for Uncertainty Quantification and Machine Learning (Final Technical Report)

With increasing simulation and measurement data, machine learning and artificial intelligence have been widely used in computational decision-making of complex engineering systems. The resulting tools, such as uncertainty quantification solvers, reinforcement learning, and physics-informed machine learning, have achieved great success in critical DOE tasks such as material discovery and design, energy system modeling and control, and numerical weather and climate prediction. A core topic in scientific machine learning and artificial intelligence is Bayesian inference: given an observed data set, people want to estimate the posterior distribution of a (possibly large) number of hidden parameters. Due to the flexibility and weak assumptions, Bayesian sampling has been the mainstream Bayesian inference solvers despite the rapid progress of approximate Bayesian inference. Classical Bayesian sampling methods such as Markov-chain Monte Carlo suffer from a low-acceptance rate due to the random walk nature, therefore state-of-the-art techniques use Hamiltonian Monte Carlo and its variants to efficiently draw posterior samples in a high dimension. The key idea of Hamiltonian Monte Carlo and its variants is to simulate the Hamiltonian dynamics of a classical particle with a fixed mass, and their performance significantly degrades when the posterior distribution is highly spiky or has multiple modes. Leveraging the idea of quantum physics, this project has investigated new theory, algorithms and applications of Bayesian inference (especially Bayesian sampling). The main results include: (1) novel quantum-inspired Bayesian sampling methods that can lead to better accuracy for challenging multi-modal or spiky distributions, (2) more scalable machine learning framework leveraging tensor-compressed Bayesian inference, and (3) Bayesian and sampling approaches for verifying the robustness of continuous and binary neural networks.

97 MATHEMATICS AND COMPUTING↗