Search NASA⌕ Search

SEARCH · Search NASA

Results for “Gaussian process learning”

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 451 records · Page 25
All sourcesNASA NTRSDOE OSTI
DOE OSTI2019

A dynamic data-driven method for dealing with model structural error in soil moisture data assimilation

Attributing to the flexibility in considering various types of observation error and model error, data assimilation has been increasingly applied to dynamically improve soil moisture modeling in many hydrological practices. However, accurate characterization of model error, especially the part caused by defective model structure, presents a significant challenge to the successful implementation of data assimilation. Model structural error has received limited attention relative to parameter and input errors, mainly due to our poor understanding of structural inadequacy and the difficulties in parameterizing structural error. In this paper, we present a dynamic data-driven approach to estimate the model structural error in soil moisture data assimilation without the need for identifying error generation mechanism or specifying particular form for the error model. The error model is based on the Gaussian process regression and then integrated into the ensemble Kalman filter (EnKF) to form a hybrid method for dealing with multi-source model errors. Two variants of the hybrid method in terms of two different error correction manners are proposed. The effectiveness of the proposed method is tested through a suit of synthetic cases and a real-world case. Results demonstrate the potential of the proposed hybrid method for estimating model structural error and providing improved model predictions. Compared to the traditional EnKF without explicitly considering the model structural error, parameter compensation issue is obviously reduced and soil moisture retrieval is substantially improved.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2024

Addressing Experimental Challenges in Probing Dark Energy with Accuracy and Precision with the Rubin Observatory Legacy Survey of Space & Time (LSST)

With support from this award, the Stanford group under the leadership of PI Patricia Burchat focused on pixel-level algorithms that are key to the investigation of dark energy with the Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST). The work was carried out through strong engagement in the DOE-supported LSST Dark Energy Science Collaboration (DESC), for which the PI served as Deputy Spokesperson (07/2019 - 07/2021) and Technical Coordinator (07/2021 - 07/2023). Members of the Burchat group are Full DESC members and are among scientists who have been accepted to join the Rubin Observatory System Integration, Test, and Commissioning (SIT-Com) effort. The specific intellectual focus of the group was the optimization of gravitational lensing as a probe of the evolution of the distribution of dark matter in space and time to learn about the nature of dark energy. The Stanford group pursued an integrated research program to understand in detail, and then remediate or calibrate, systematic effects that could limit the sensitivity of the LSST to cosmic shear as a dark energy probe -- particularly those related to our understanding of the atmospheric point spread function (PSF), challenges associated with blended galaxy images, and impacts of residual camera artifacts. Key to this work was developing, enhancing, and validating the fidelity of simulations that play a key role in the development of our understanding of subtle systematic biases and then using these simulations to develop techniques to characterize and mitigate biases, and associated algorithms for implementation in the LSST or DESC science pipelines. The group’s work included contributions to the open-source GalSim simulation package, which is widely used in the community; simulations and studies of spatial correlations of atmospheric PSFs; studies of temporal correlations of the atmospheric PSF; impacts of chromatic PSFs on cosmic shear calibration; continued studies of anisotropic Gaussian process interpolation of PSF models across the focal plane; and measuring and mitigating impacts of residual camera artifacts on cosmic shear. A primary impact of this work is that it increases the scientific discovery potential of the DOE and NSF investments in the Rubin Observatory Legacy Survey of Space \& Time, by addressing challenges in interpreting astronomical images through more accurate descriptions of the blurring of images due to the atmosphere. Specifically, this work impacts our ability to understand the fundamental nature of dark energy.

Burchat, Patricia Rose↗
DOE OSTI2022

Learning the temporal evolution of multivariate densities via normalizing flows

In this work, we propose a method to learn multivariate probability distributions using sample path data from stochastic differential equations. Specifically, we consider temporally evolving probability distributions (e.g., those produced by integrating local or nonlocal Fokker–Planck equations). Here, we analyze this evolution through machine learning assisted construction of a time-dependent mapping that takes a reference distribution (say, a Gaussian) to each and every instance of our evolving distribution. If the reference distribution is the initial condition of a Fokker–Planck equation, what we learn is the time-T map of the corresponding solution. Specifically, the learned map is a multivariate normalizing flow that deforms the support of the reference density to the support of each and every density snapshot in time. We demonstrate that this approach can approximate probability density function evolutions in time from observed sampled data for systems driven by both Brownian and Lévy noise. We present examples with two- and three-dimensional, uni- and multimodal distributions to validate the method.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2016

ELM: AN ALGORITHM TO ESTIMATE THE ALPHA ABUNDANCE FROM LOW-RESOLUTION SPECTRA

We have investigated a novel methodology using the extreme learning machine (ELM) algorithm to determine the α abundance of stars. Applying two methods based on the ELM algorithm—ELM+spectra and ELM+Lick indices—to the stellar spectra from the ELODIE database, we measured the α abundance with a precision better than 0.065 dex. By applying these two methods to the spectra with different signal-to-noise ratios (S/Ns) and different resolutions, we found that ELM+spectra is more robust against degraded resolution and ELM+Lick indices is more robust against variation in S/N. To further validate the performance of ELM, we applied ELM+spectra and ELM+Lick indices to SDSS spectra and estimated α abundances with a precision around 0.10 dex, which is comparable to the results given by the SEGUE Stellar Parameter Pipeline. We further applied ELM to the spectra of stars in Galactic globular clusters (M15, M13, M71) and open clusters (NGC 2420, M67, NGC 6791), and results show good agreement with previous studies (within 1σ). A comparison of the ELM with other widely used methods including support vector machine, Gaussian process regression, artificial neural networks, and linear least-squares regression shows that ELM is efficient with computational resources and more accurate than other methods.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2023

Localized keyhole pore prediction during laser powder bed fusion via multimodal process monitoring and X-ray radiography

Systematic fault detection and control during laser powder bed fusion (L-PBF) has been a long-standing objective for system manufacturers and researchers in the additive manufacturing (AM) industry. This manuscript investigates a data fusion approach for detection of keyhole porosity formation during laser irradiation of Ti-6Al-4V substrates by concurrent recording of thermally induced optical emission measured using both off-axis and coaxial photodiode sensors, and acoustic emission. Subsurface defect formation was monitored via high-speed synchrotron X-ray imaging at 20,000 frames per second, enabling temporal registration of keyhole pore formation events to the monitoring signals at a resolution of 50 µs. We developed data fusion machine learning (ML) models for localized prediction of keyhole pore formation at various time scales ranging from 0.5 ms to 2 ms. The signal segments were featurized using two independent approaches: (1) power spectral density (PSD) and (2) highly comparative time series analysis (HCTSA) framework. The extracted features from different sensor modalities were fused together to construct a multimodal feature space and sequential feature selection was used to determine the most informative features for training the ML models. The predictive performance was evaluated for three classifying algorithms: Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Gaussian Naive Bayes (GNB). As a result, pore formation events were predicted with up to 0.95 F1-score, 1.0 recall and 0.94 accuracy. The most heavily weighted features indicate that model performance is chiefly governed by the acoustic monitoring signal, with a secondary contribution from the optical emission sensors.

36 MATERIALS SCIENCE↗
DOE OSTI2015

BAYESIAN CALIBRATION OF SAFETY CODES USING DATA FROM SEPARATE-AND INTEGRAL EFFECTS TESTS

Large-scale system codes for simulation of safety performance of nuclear plants may contain parameters whose values are not known very accurately. In order to be able to use the results of these simulation codes with confidence, it is important to learn how the uncertainty on the values of these parameters affects the output of the codes. New information from tests or operating experience is incorporated into safety codes by a process known as calibration, which reduces uncertainty in the output of the safety code, and thereby improves its support for decision-making. Modern analysis capabilities afford very significant improvements on classical ways of doing calibration, and the work reported here implements some of those improvements. The key innovation has come from development of safety code surrogate model (code emulator) construction and prediction algorithms. A surrogate is needed for calibration of plant-scale simulation codes because the multivariate nature of the problem (i.e., the need to adjust multiple uncertain parameters at once to fit multiple pieces of new information) calls for multiple evaluations of performance, which, for a computation-intensive model, makes calibration very computation-intensive. Use of a fast surrogate makes the calibration processes used here with Markov Chain Monte Carlo (MCMC) sampling feasible. Moreover, most traditional surrogates do not provide uncertainty information along with their predictions, but the Gaussian Process (GP) based code surrogates used here do. This improves the soundness of the code calibration process. Results are demonstrated on a simplified scenario with data from Separate and Integral Effect Tests.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2025

Machine learning inversion from small-angle scattering for charged polymers

We develop Monte Carlo simulations for uniformly charged polymers and a machine learning algorithm to interpret the intra-polymer structure factor of the charged polymer system, which can be obtained from small-angle scattering experiments. The polymer is modeled as a chain of fixed-length bonds, where the connected bonds are subject to bending energy, and there is also a screened Coulomb potential for charge interaction between all joints. The bending energy is determined by the intrinsic bending stiffness, and the charge interaction depends on the interaction strength and screening length. All three contribute to the stiffness of the polymer chain and lead to longer and larger polymer conformations. The screening length also introduces a second length scale for the polymer besides the bending persistence length. To obtain the inverse mapping from the structure factor to these polymer conformation and energy-related parameters, we generate a large data set of structure factors by running simulations for a wide range of polymer energy parameters. We use principal component analysis to investigate the intra-polymer structure factors and determine the feasibility of the inversion using the nearest neighbor distance. We employ Gaussian process regression to achieve the inverse mapping and extract the characteristic parameters of polymers from the structure factor with low relative error.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Deep Learning of Dark Energy Spectroscopic Instrument Mock Spectra to Find Damped Lyα Systems

We have updated and applied a convolutional neural network (CNN) machine-learning model to discover and characterize damped Ly α systems (DLAs) based on Dark Energy Spectroscopic Instrument (DESI) mock spectra. We have optimized the training process and constructed a CNN model that yields a DLA classification accuracy above 99% for spectra that have signal-to-noise ratios (S/N) above 5 per pixel. The classification accuracy is the rate of correct classifications. This accuracy remains above 97% for lower S/N ≈1 spectra. This CNN model provides estimations for redshift and H i column density with standard deviations of 0.002 and 0.17 dex for spectra with S/N above 3 pixel -1 . Also, this DLA finder is able to identify overlapping DLAs and sub-DLAs. Further, the impact of different DLA catalogs on the measurement of baryon acoustic oscillations (BAO) is investigated. The cosmological fitting parameter result for BAO has less than 0.61% difference compared to analysis of the mock results with perfect knowledge of DLAs. This difference is lower than the statistical error for the first year estimated from the mock spectra: above 1.7%. We also compared the performances of the CNN and Gaussian Process (GP) models. Our improved CNN model has moderately 14% higher purity and 7% higher completeness than an older version of the GP code, for S/N > 3. Both codes provide good DLA redshift estimates, but the GP produces a better column density estimate by 24% less standard deviation. A credible DLA catalog for the DESI main survey can be provided by combining these two algorithms.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2026

Describing Point Defect Topology in 2D Energy Materials Through Computer Vision

Point defects such as vacancies and impurity atoms strongly impact the performance of 2D materials. Traditional efforts often rely on manual detection, a process that is time-intensive, prone to human error, and challenging to scale. Here we leverage machine learning (ML) methods to identify and quantify vacancies within 2D transition metal carbides (Ti3C2, MXenes), aiming to expedite detection while improving accuracy. MXenes exhibit valuable defect-defined electrochemical properties, but we currently lack statistical understanding of defect topology needed to fully harness these materials. Here we employ a convolutional neural network for semantic segmentation of experimental MXene images, opening an opportunity to conduct a rigorous statistical study on defect hierarchy while investigating local relaxation in the lattice. We show how the integration of ML can yield fundamental insight into point defects, providing a powerful tool that will play an increasingly crucial role in the future of materials science. ML is often not just a matter of straightforward application, and pretrained models proved ineffective in this case. Instead, we trained our own neural network (NN) and applied data augmentation techniques and fine-tuning to the training dataset. Since labeled microscopy data is often scarce, we developed training data from a previously published wide-frame MXene image, using customized Gaussian fitting to locate atomic positions. Our trained model was then applied to a large dataset of experimental images, enabling a statistical study of defect configurations across three samples prepared with different HF etchant concentrations (5%, 9.1%, and 12.5%), as shown in Fig. 1. This also allowed us to investigate local strain around vacancies, though we find that we are limited by the precision of measurements using high-angle annular dark field (HAADF) images, as shown in Fig. 2. This study demonstrates how ML enables large-scale, quantitative analysis of atomic defects - an otherwise infeasible task with traditional methods. While our NN was specialized for Ti3C2 MXenes, the pipeline we developed provides a foundation for future ML models tailored to other materials. Ultimately, we envision embedding the NN onto the microscope to give real-time feedback to the user. To make this a reality, continued work is necessary to fully understand the NN's capabilities and limitations. This study gets one step closer to our goals of automated experimentation moving away from traditional methods of manual labeling. As ML capabilities advance, we hope to continue adapting and applying these techniques in microscopy.

2D materials↗
DOE OSTI2024

Portable, heterogeneous ensemble workflows at scale using libEnsemble

libEnsemble is a Python-based toolkit for running dynamic ensembles, developed as part of the DOE Exascale Computing Project. The toolkit utilizes a unique generator–simulator–allocator paradigm, where generators produce input for simulators, simulators evaluate those inputs, and allocators decide whether and when a simulator or generator should be called. The generator steers the ensemble based on simulation results. Generators may, for example, apply methods for numerical optimization, machine learning, or statistical calibration. libEnsemble communicates between a manager and workers. Flexibility is provided through multiple manager–worker communication substrates each of which has different benefits. These include Python’s multiprocessing, mpi4py, and TCP. Multisite ensembles are supported using Balsam or Globus Compute. We overview the unique characteristics of libEnsemble as well as current and potential interoperability with other packages in the workflow ecosystem. We highlight libEnsemble’s dynamic resource features: libEnsemble can detect system resources, such as available nodes, cores, and GPUs, and assign these in a portable way. These features allow users to specify the number of processors and GPUs required for each simulation; and resources will be automatically assigned on a wide range of systems, including Frontier, Aurora, and Perlmutter. Such ensembles can include multiple simulation types, some using GPUs and others using only CPUs, sharing nodes for maximum efficiency. We also describe the benefits of libEnsemble’s generator–simulator coupling, which easily exposes to the user the ability to cancel, and portably kill, running simulations based on models that are updated with intermediate simulation output. We demonstrate libEnsemble’s capabilities, scalability, and scientific impact via a Gaussian process surrogate training problem for the longitudinal density profile at the exit of a plasma accelerator stage. In conclusion, the study uses gpCAM for the surrogate model and employs either Wake-T or WarpX simulations, highlighting efficient use of resources that can easily extend to exascale.

Dynamic ensembles↗
DOE OSTI2024

Forward variable selection enables fast and accurate dynamic system identification with Karhunen-Loève decomposed Gaussian processes

A promising approach for scalable Gaussian processes (GPs) is the Karhunen-Loève (KL) decomposition, in which the GP kernel is represented by a set of basis functions which are the eigenfunctions of the kernel operator. Such decomposed kernels have the potential to be very fast, and do not depend on the selection of a reduced set of inducing points. However KL decompositions lead to high dimensionality, and variable selection thus becomes paramount. This paper reports a new method of forward variable selection, enabled by the ordered nature of the basis functions in the KL expansion of the Bayesian Smoothing Spline ANOVA kernel (BSS-ANOVA), coupled with fast Gibbs sampling in a fully Bayesian approach. It quickly and effectively limits the number of terms, yielding a method with competitive accuracies, training and inference times for tabular datasets of low feature set dimensionality. Theoretical computational complexities are O ( N P 2 ) in training and O ( P ) per point in inference, where N is the number of instances and P the number of expansion terms. The inference speed and accuracy makes the method especially useful for dynamic systems identification, by modeling the dynamics in the tangent space as a static problem, then integrating the learned dynamics using a high-order scheme. The methods are demonstrated on two dynamic datasets: a ‘Susceptible, Infected, Recovered’ (SIR) toy problem, along with the experimental ‘Cascaded Tanks’ benchmark dataset. Comparisons on the static prediction of time derivatives are made with a random forest (RF), a residual neural network (ResNet), and the Orthogonal Additive Kernel (OAK) inducing points scalable GP, while for the timeseries prediction comparisons are made with LSTM and GRU recurrent neural networks (RNNs) along with the SINDy package.

Hayes, Kyle↗
DOE OSTI2023

Exact Gaussian processes for massive datasets via non-stationary sparsity-discovering kernels

Abstract A Gaussian Process (GP) is a prominent mathematical framework for stochastic function approximation in science and engineering applications. Its success is largely attributed to the GP’s analytical tractability, robustness, and natural inclusion of uncertainty quantification. Unfortunately, the use of exact GPs is prohibitively expensive for large datasets due to their unfavorable numerical complexity of $$O(N^3)$$ O ( N 3 ) in computation and $$O(N^2)$$ O ( N 2 ) in storage. All existing methods addressing this issue utilize some form of approximation—usually considering subsets of the full dataset or finding representative pseudo-points that render the covariance matrix well-structured and sparse. These approximate methods can lead to inaccuracies in function approximations and often limit the user’s flexibility in designing expressive kernels. Instead of inducing sparsity via data-point geometry and structure, we propose to take advantage of naturally-occurring sparsity by allowing the kernel to discover—instead of induce—sparse structure. The premise of this paper is that the data sets and physical processes modeled by GPs often exhibit natural or implicit sparsities, but commonly-used kernels do not allow us to exploit such sparsity. The core concept of exact, and at the same time sparse GPs relies on kernel definitions that provide enough flexibility to learn and encode not only non-zero but also zero covariances. This principle of ultra-flexible, compactly-supported, and non-stationary kernels, combined with HPC and constrained optimization, lets us scale exact GPs well beyond 5 million data points.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

A surrogate-model-based approach for estimating the first and second-order moments of offshore wind power

Power curve, the functional relationship that governs the process of converting a set of weather variables experienced by a wind turbine into electric power, is widely used in the wind industry to estimate power output for planning and operational purposes. Existing methods for power curve estimation have three main limitations: (i) they mostly rely on wind speed as the sole input, thus ignoring the secondary, yet possibly significant effects of other environmental factors, (ii) they largely overlook the complex marine environment in which offshore turbines operate, potentially compromising their value in offshore wind energy applications, and (ii) they solely focus on the first-order properties of wind power, with little (or null) information about the variation around the mean behavior, which is important for ensuring reliable grid integration, asset health monitoring, and energy storage, among others. In light of that, this study investigates the impact of several wind-and wave-related factors on offshore wind power variability, with the ultimate goal of accurately predicting its first two moments. Further, our approach couples OpenFAST—a multi-physics wind turbine simulator—with Gaussian Process (GP) regression to reveal the underlying relationships governing offshore weather-to-power conversion. We first find that a multi-input power curve which captures the combined impact of wind speed, direction, and air density, can provide double-digit improvements, in terms of prediction accuracy, relative to univariate methods which rely on wind speed as the sole explanatory variable (e.g. the standard method of bins). Wave-related variables are found not important for predicting the average power output, but interestingly, appear to be extremely relevant in describing the fluctuation of the offshore power around its mean. Tested on real-world data collected at the New York/New Jersey bight, our proposed multi-input models demonstrate a high explanatory power in predicting the first two moments of offshore wind generation, testifying their potential value to the offshore wind industry.

17 WIND ENERGY↗
DOE OSTI2024

A novel conditional generative model for efficient ensemble forecasts of state variables in large-scale geological carbon storage

Integrating monitoring data to efficiently update reservoir pressure and CO 2 plume distribution forecasts presents a significant challenge in geological carbon storage (GCS) applications. Inverse modeling techniques are commonly used to fuse observational data and refine reservoir model parameters, thereby improving state variable forecasts. However, these techniques often rely on linear or Gaussian assumptions, which can limit their effectiveness in accurately predicting state variables. Moreover, simulating large-scale three-dimensional (3D) GCS problems is computationally expensive, making iterative runs in inverse problems prohibitive. To address these challenges, we propose a conditional generative model utilizing the score-based diffusion method for real-time 3D pressure and saturation field distribution predictions. Our approach involves solving the score function with a mini-batch-based Monte Carlo estimator to generate labeled data. This data is subsequently employed to train a fully connected neural network, enabling it to learn the conditional sample generator within a supervised learning framework. This method enables the rapid generation of a large ensemble of predictions, facilitating comprehensive uncertainty quantification of state variables. Here we applied our method to forecast the dynamic 3D distributions of pressure and saturation fields over a 30-year injection period. The statistical assessment with low root mean square error (RMSE) values demonstrates that our method can accurately predict the spatiotemporal distributions of both pressure and saturation fields. Moreover, the developed conditional generative model shows high computational efficiency by generating 100 ensemble forecasts of 3D state variables in less than 10 min. The consistency between ensemble averages and ground truth values further illustrates the model’s capability to capture state variable dynamics during the CO 2 plume injection process. Notably, the ground truth values fall within the ensemble forecasts, indicating that our uncertainty quantification effectively captures variability and potential noise in the observations. Thus, the developed conditional generative model proves to be a more efficient, accurate, and practical tool for GCS applications, facilitating timely risk analysis and informed decision-making.

58 GEOSCIENCES↗
DOE OSTI2016

Next Day Building Load Predictions based on Limited Input Features Using an On-Line Laterally Primed Adaptive Resonance Theory Artificial Neural Network.

Optimal integration of thermal energy storage within commercial building applications requires accurate load predictions. Several methods exist that provide an estimate of a buildings future needs. Methods include component-based models and data-driven algorithms. This work implemented a previously untested algorithm for this application that is called a Laterally Primed Adaptive Resonance Theory (LAPART) artificial neural network (ANN). The LAPART algorithm provided accurate results over a two month period where minimal historical data and a small amount of input types were available. These results are significant, because common practice has often overlooked the implementation of an ANN. ANN have often been perceived to be too complex and require large amounts of data to provide accurate results. The LAPART neural network was implemented in an on-line learning manner. On-line learning refers to the continuous updating of training data as time occurs. For this experiment, training began with a singe day and grew to two months of data. This approach provides a platform for immediate implementation that requires minimal time and effort. The results from the LAPART algorithm were compared with statistical regression and a component-based model. The comparison was based on the predictions linear relationship with the measured data, mean squared error, mean bias error, and cost savings achieved by the respective prediction techniques. The results show that the LAPART algorithm provided a reliable and cost effective means to predict the building load for the next day.

25 ENERGY STORAGE↗
DOE OSTI2018

Efficient Surrogate Model Development: Impact of Sample Size and Underlying Model Dimensions

This study compares eight surrogate-model construction approaches using computational experiments. Here, the construction approaches considered include: Artificial Neural Networks (ANNs), Automated Learning of Algebraic Models using Optimization (ALAMO), Radial Basis Networks (RBNs), Extreme Learning Machines (ELMs), Gaussian Progress Regression (GPR), Random Forests (RFs), Support Vector Regression (SVR), and Multivariate Adaptive Regression Splines. Each approach is used to construct surrogate models for predicting the outputs of thirty-four test functions. The data used to construct surrogate models were generated using Latin Hypercube Sampling (LHS), Halton and Sobol sampling methods. We studied the impact of sampling method, sample size, and challenge function characteristics on the accuracy of surrogate model predictions. Our results revealed that for smaller sample sizes, on average, surrogate models trained using the data points generated by Sobol sequences provided the best estimation for all surrogate model construction approaches. As the sample size increased, the impact of sampling methods diminished. The comparison of surrogate-modeling construction approaches at large sample sizes revealed that surrogate models trained using ANN, ALAMO, and ELM yielded smaller root mean squared error and higher adjusted R-squared values than the models trained using the rest of the approaches.

42 ENGINEERING↗
DOE OSTI2025

Data-Efficient Methods for Determining Flory–Huggins χ Parameters in Multicomponent Polymer Formulations

Polymer formulations are essential in diverse applications including personal care products, coatings, paints, adhesives, and plastic materials. Designing these formulations requires navigating large, complex design spaces, where phase and self-assembly behavior critically impact performance. The Flory–Huggins χ parameter, which quantifies segmental miscibility, is widely used to parametrize the excess free energy of mixing in formulation models. In this work, we introduce two data-efficient, top-down methods for estimating χ parameters using the Random Phase Approximation (RPA): (i) Boundary Nonlinear Regression (Boundary-NLR), which fits theoretical spinodal boundaries to experimental phase boundaries, and (ii) Surrogate Model Inverse Parameter Estimation (SMIPE), which uses a Gaussian Process Classifier to fit sparse phase maps via a surrogate model. Both methods allow rapid parametrization of polymer field-theoretic models without the need for additional experiments. We evaluate these approaches on data sets involving polymer–solvent–nonsolvent ternary mixtures and block copolymer–solvent systems, demonstrating their robustness to experimental noise and their relevance for real-world formulation design.

copolymers↗
DOE OSTI2025

CAFE AU LAIT: Compute-Aware Federated Augmented Low-Rank AI Training

Federated finetuning is crucial for unlocking the knowledge embedded in pretrained Large Language Models (LLMs) when data are geographically distributed across clients. Unlike finetuning with data from a single institution, federated finetuning allows collaboration across multiple institutions, enabling the utilization of diverse and decentralized datasets while preserving data privacy. Given the high computing costs of LLM training and the emphasis on energy efficiency in Federated Learning (FL), Low-Rank Adaptation (LoRA) has emerged as a widely adopted algorithm due to its significantly reduced number of trainable parameters. However, this assumes that all data silos have the necessary computing resources to compute local updates of LLMs. Nevertheless, in practice, the computing resources across clients are highly heterogeneous: while some may have access to hundreds of GPUs, others might have limited or no GPU access. Recently, federated finetuning using synthetic data has been proposed, allowing clients to participate in a collaborative training run without training LLMs locally. However, our experimental results reveal a performance gap between models trained using synthetic data and those trained using local updates. Motivated by the observed heterogeneity in computing resources and the performance gap, we propose a novel two-stage algorithm that leverages the storage and computing capabilities of a strong server. In the first stage, under the coordination of the strong server, clients with limited computing resources collaborate to generate synthetic data, which is transferred to and stored on the strong server. In the second stage, the strong server uses this synthetic data on behalf of the resource-constrained clients to perform federated LoRA finetuning alongside clients with sufficient computing resources. This approach ensures that all clients can participate in the finetuning process. Experimental results demonstrate that incorporating local updates from even a small fraction of clients improves performance compared to using synthetic data for all clients. Furthermore, we incorporate the Gaussian mechanism in both stages to guarantee client-level differential privacy.

Wang, Jiayi [ORNL]↗