SEARCH · Search NASA

Results for “interpretable machine 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.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 721 records · Page 40

All sources NASA NTRS DOE OSTI

DOE OSTI2024

Evaluating the Effectiveness of Retrieval-Augmented Large Language Models in Scientific Document Reasoning

Despite the dramatic progress in Large Language Model (LLM) development, LLMs often provide seemingly plausible but not factual information, often referred as hallucinations. Retrieval-augmented LLMs provide a non-parametric approach to solve these issues by retrieving relevant information from external data sources and augment the training process. These models helps to trace evidence from an externally provided knowledge base allowing the model predictions to be better interpreted and verified. In this work, we critically evaluate these models in their ability to perform in scientific document reasoning tasks. To this end, we tuned multiple such model variants with science-focused instructions and evaluated them on a scientific document reasoning benchmark for the usefulness of the retrieved document passages. Our findings suggest that models justify predictions in science tasks with fabricated evidence and leveraging scientific corpus as pretraining data does not alleviate the risk of evidence fabrication.

• Artificial intelligence (AI) / machine learning ↗

DOE OSTI2025

AI Benchmark Democratization and Carpentry

Benchmarks are a cornerstone of modern machine learning, enabling reproducibility, comparison, and scientific progress. However, AI benchmarks are increasingly complex, requiring dynamic, AI-focused workflows. Rapid evolution in model architectures, scale, datasets, and deployment contexts makes evaluation a moving target. Large language models often memorize static benchmarks, causing a gap between benchmark results and real-world performance. Beyond traditional static benchmarks, continuous adaptive benchmarking frameworks are needed to align scientific assessment with deployment risks. This calls for skills and education in AI Benchmark Carpentry. From our experience with MLCommons, educational initiatives, and programs like the DOE's Trillion Parameter Consortium, key barriers include high resource demands, limited access to specialized hardware, lack of benchmark design expertise, and uncertainty in relating results to application domains. Current benchmarks often emphasize peak performance on top-tier hardware, offering limited guidance for diverse, real-world scenarios. Benchmarking must become dynamic, incorporating evolving models, updated data, and heterogeneous platforms while maintaining transparency, reproducibility, and interpretability. Democratization requires both technical innovation and systematic education across levels, building sustained expertise in benchmark design and use. Benchmarks should support application-relevant comparisons, enabling informed, context-sensitive decisions. Dynamic, inclusive benchmarking will ensure evaluation keeps pace with AI evolution and supports responsible, reproducible, and accessible AI deployment. Community efforts can provide a foundation for AI Benchmark Carpentry.

von Laszewski, Gregor [Virginia U.]↗

DOE OSTI2023

Harnessing Neural Networks for Elucidating X-ray Absorption Structure–Spectrum Relationships in Amorphous Carbon

Improved understanding of structural and chemical properties through local experimental probes, such as X-ray absorption near-edge structure (XANES) spectroscopy, is crucial for the understanding and design of functional materials. In recent years, significant advancements have been made in the development of data science approaches for the automated interpretation of XANES structure–spectrum relationships. However, existing studies have primarily focused on crystalline solids and small molecules, while fewer efforts have been devoted to disordered systems. Thus, in this work, we demonstrate the development of neural network models for predicting and interpreting XANES spectra of amorphous carbon (a-C) from local structural descriptors. Comparison between different structural descriptors expectedly shows that the inclusion of both bond length and bond angle information is necessary for an accurate prediction of the spectra. Among the descriptors considered in this work, we find that the local many-body tensor representation yields the highest accuracy and greatest interpretability so that it can be leveraged to understand the importance of structural motifs in determining XANES spectra. Furthermore, we also discuss performance of neural network models for predicting both local structure features, such as bond lengths and bond angles, and global chemical composition, such as the sp:sp 2 :sp 3 ratio.

36 MATERIALS SCIENCE↗

DOE OSTI2021

Performance of Pulsed Thermal Tomography Imaging with Machine Learning-Based Classification of Defects in Additively Manufactured Structures

Additive manufacturing (AM) is an emerging method for cost-efficient fabrication of complex topology nuclear reactor parts from high-strength corrosion resistance alloys, such as stainless steel and Inconel. AM of metallic structures for nuclear energy applications is currently based on laser powder bed fusion (LPBF) process. Some of the challenges with using LPBF method for nuclear manufacturing include the possibility of introducing pores into metallic structures. Integrity of AM structures needs to be evaluated nondestructively because material flaws could lead to premature failures in high temperature nuclear reactor environment. Currently, there exist limited capabilities to evaluate actual AM structures non-destructively. Pulsed Thermal Tomography Imaging (PTT) provides a capability for non-destructive evaluation (NDE) of subsurface defects in arbitrary size structures. The PTT method is based on recording material surface temperature transients with infrared (IR) camera following thermal pulse delivered on material surface with flash light. The PTT method has advantages for NDE of actual AM structures because the method involves one-sided non-contact measurements and fast processing of large sample areas captured in one image. Following initial qualification of an AM component for deployment in a nuclear reactor, a PTT system can also be used for in-service nondestructive evaluation (NDE) applications. In this report, we describe recent progress in enhancing PTT capabilities in detecting and visualizing microscopic defects in metallic specimens. The thermal tomography (TT) algorithm obtains depth reconstructions of spatial effusivity from the data cube of sequentially recorded surface temperatures. However, interpretation of TT images is non-trivial because of blurring of images with increasing depth. To address this challenge, we have developed a deep learning convolutional neural network (CNN) to classify size and orientation subsurface defects in simulated TT images. CNN is trained on a database of TT images created for a set of simulated metallic structures with elliptical subsurface voids. Test of CNN performance demonstrate the ability to classify radii and angular orientation of subsurface defects in TT images. In addition, we have shown that CNN trained on elliptical defects is capable of classifying irregular-shaped defects obtained from scanning electron microscopy (SEM) of stainless steel sections printed with LPBF.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

DOE OSTI2022

Machine learning for design principles for single atom catalysts towards electrochemical reactions

Machine learning (ML) integrated density functional theory (DFT) calculations have recently been used to accelerate the design and discovery of heterogeneous catalysts such as single atom catalysts (SACs) through the establishment of deep structure–activity relationships. Here, this review provides recent progress in the ML-aided rational design of heterogeneous catalysts with the focus on SACs in terms of structure–activity relationships, feature importance analysis, high-throughput screening, stability, and metal–support interactions for electrochemistry. Support vector machine (SVM), random forest regression (RFR), and deep neural networks (DNN) along with atomic properties are mainly used for the design of SACs. The ML results have shown that the number of electrons in the d orbital, oxide formation enthalpy, ionization energy, Bader charge, d-band center, and enthalpy of vaporization are mainly the most important parameters for the defining of the structure–activity relationships for electrochemistry. However, the black-box nature of ML techniques occasionally makes a physical interpretation of descriptors, such as the Bader charge, d-band center, and enthalpy of vaporization, non-trivial. At the current stage, ML application is limited by the lack of a large and high-quality database. Future prospects for the development of a large database and a generalized ML algorithm for SAC design are discussed to give insights for further studies in this field.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

DOE OSTI2021

A novel construct for scaling groundwater–river interactions based on machine-guided hydromorphic classification

Hydrologic exchange between river channels and adjacent subsurface environments is a key process that influences water quality and ecosystem function in river corridors. Predictive numerical models are needed to understand responses of river corridors to environmental change and to support sustainable watershed management. We posit that systematic hydromorphic classification provides a scaling construct that facilitates extrapolation of outputs from local-scale mechanistic models to reduced-order models applicable at reach and watershed scales. This in turn offers the potential to improve large-scale predictions of river corridor hydrobiogeochemical processes. Here we present a new machine-guided hydromorphic classification methodology that addresses the key requirements of this objective, and we demonstrate its application to a segment of the Columbia River in the northwestern United States. The resulting hydromorphic classes form spatially coherent and physically interpretable hydromorphic units that exhibit distinct behaviors in terms of distributions of subsurface residence times (a primary control on critical biogeochemical reactions). This approach forms the basis of ongoing research that is evaluating the formulation of reduced-order models and transferability of results to other river reaches and larger scales.

54 ENVIRONMENTAL SCIENCES↗

DOE OSTI2023

Integrating LEO and GEO Observations: Toward Optimal Summertime Satellite Precipitation Retrieval

Abstract Reliable quantitative precipitation estimation with a rich spatiotemporal resolution is vital for understanding the Earth’s hydrological cycle. Precipitation estimation over land and coastal regions is necessary for addressing the high degree of spatial heterogeneity of water availability and demand, and for resolving the extremes that modulate and amplify hazards such as flooding and landslides. Advancements in computation power along with unique high spatiotemporal and spectral resolution data streams from passive meteorological sensors aboard geosynchronous Earth-orbiting (GEO) and low Earth-orbiting (LEO) satellites offer exciting opportunities to retrieve information about surface precipitation phenomena using data-driven machine learning techniques. In this study, the capabilities of U-Net–like architecture are investigated to map instantaneous, summertime surface precipitation intensity at the spatial resolution of 2 km. The calibrated brightness temperature products from the Global Precipitation Measurement (GPM) Microwave Imager (GMI) radiometer are combined with multispectral images (visible, near-infrared, and infrared bands) from the Advanced Baseline Imager (ABI) aboard the GOES-R satellites as main inputs to the U-Net–like precipitation algorithm. Total precipitable water and 2-m temperature from the Global Forecast System (GFS) model are also used as auxiliary inputs to the model. The results show that the U-Net–like algorithm can capture fine-scale patterns and intensity of surface precipitation at high spatial resolution over stratiform and convective precipitation regimes. The evaluations reveal the potential of extracting relevant, high spatial features over complex surface types such as mountainous regions and coastlines. The algorithm allows users to interpret the inputs’ importance and can serve as a starting point for further exploration of precipitation systems within the field of hydrometeorology.

Meteorology & Atmospheric Sciences↗

DOE OSTI2026

Digitizing Today’s Buildings in the Real World: Lessons from Field Demonstrations

Digital twins, created by generating a virtual replica of a building, enable safe evaluation of operational scenarios and applications like fault detection and diagnosis and advanced controls. However, a prerequisite is the creation of a machine-readable digital representation of a building, currently hindered by fragmented information scattered across mechanical drawings, point lists, and natural language sequences. As a result, digital twin development remains labor-intensive, error-prone, and difficult to validate. To address these challenges, two efforts from ASHRAE aim to support the digitalization of buildings. ASHRAE s223 establishes a semantic model of buildings, representing system components, configuration, and data sources. ASHRAE s231 defines a vendor-neutral programming language for expressing their control logic. As the industry evaluates implementing them in their products, understanding the challenges that vendors and implementers may face is crucial. In this paper, we present findings and lessons learned from field demonstrations in five buildings that implemented control applications using ASHRAE s223 and s231. The demonstrations highlight how semantic modeling and formalized control descriptions can significantly reduce software development time, manual point mapping, and hard-coding. Beyond time efficiency, they enable reliable automation by minimizing human interpretation and providing a means for consistency across projects. We describe the processes and best practices for model creation and model usage, from translating heterogeneous building documentation into semantic representations to implementing control logic in real-world systems. Finally, we discuss the challenges that persist, including integration with legacy software environments, gaps in interoperability, and the level of expertise still required to effectively leverage semantic models.

Prakash, Anand Krishnan↗

DOE OSTI2024

Jacobian-scaled K-means clustering for physics-informed segmentation of reacting flows

This work introduces Jacobian-scaled K-means (JSK-means) clustering, which is a physicsinformed clustering strategy centered on the K-means framework. The method allows for the injection of underlying physical knowledge into the clustering procedure through a distance function modification: instead of leveraging conventional Euclidean distance vectors, the JSKmeans procedure operates on distance vectors scaled by matrices obtained from dynamical system Jacobians evaluated at the cluster centroids. The goal of this work is to show how the JSKmeans algorithm - without modifying the input dataset - produces clusters that capture regions of dynamical similarity, in that the clusters are redistributed towards high-sensitivity regions in phase space and are described by similarity in the source terms of samples instead of the samples themselves. The algorithm is demonstrated on a complex reacting flow simulation dataset (a channel detonation configuration), where the dynamics in the thermochemical composition space are known through the highly nonlinear and stiff Arrhenius-based chemical source terms. Interpretations of cluster partitions in both physical space and composition space reveal how JSK-means shifts clusters produced by standard K-means towards regions of high chemical sensitivity (e.g., towards regions of peak heat release rate near the detonation reaction zone). Furthermore, the findings presented here illustrate the benefits of utilizing Jacobian-scaled distances in clustering techniques, and the JSK-means method in particular displays promising potential for improving former partition-based modeling strategies in reacting flow (and other multi-physics) applications.

Clustering↗

DOE OSTI2020

Deep neural network improves the estimation of polygenic risk scores for breast cancer

Polygenic risk scores (PRS) estimate the genetic risk of an individual for a complex disease based on many genetic variants across the whole genome. Here, we compared a series of computational models for estimation of breast cancer PRS. A deep neural network (DNN) was found to outperform alternative machine learning techniques and established statistical algorithms, including BLUP, BayesA, and LDpred. In the test cohort with 50% prevalence, the Area Under the receiver operating characteristic Curve (AUC) were 67.4% for DNN, 64.2% for BLUP, 64.5% for BayesA, and 62.4% for LDpred. BLUP, BayesA, and LPpred all generated PRS that followed a normal distribution in the case population. However, the PRS generated by DNN in the case population followed a bimodal distribution composed of two normal distributions with distinctly different means. This suggests that DNN was able to separate the case population into a high-genetic-risk case subpopulation with an average PRS significantly higher than the control population and a normal-genetic-risk case subpopulation with an average PRS similar to the control population. This allowed DNN to achieve 18.8% recall at 90% precision in the test cohort with 50% prevalence, which can be extrapolated to 65.4% recall at 20% precision in a general population with 12% prevalence. Interpretation of the DNN model identified salient variants that were assigned insignificant p values by association studies, but were important for DNN prediction. These variants may be associated with the phenotype through nonlinear relationships.

59 BASIC BIOLOGICAL SCIENCES↗

DOE OSTI2026

Identifying Anomalous DESI Galaxy Spectra with a Variational Autoencoder

The tens of millions of spectra being captured by the Dark Energy Spectroscopic Instrument (DESI) provide tremendous discovery potential. In this work we show how Machine Learning, in particular Variational Autoencoders (VAE), can detect anomalies in a sample of approximately 200,000 DESI spectra comprising galaxies, quasars and stars. We demonstrate that the VAE can compress the dimensionality of a spectrum by a factor of 100, while still retaining enough information to accurately reconstruct spectral features. We then detect anomalous spectra as those with high reconstruction error and those which are isolated in the VAE latent representation. The anomalies identified fall into two categories: spectra with artefacts and spectra with unique physical features. Awareness of the former can help to improve the DESI spectroscopic pipeline; whilst the latter can lead to the identification of new and unusual objects. To further curate the list of outliers, we use the Astronomaly package which employs Active Learning to provide personalised outlier recommendations for visual inspection. In this work we also explore the VAE latent space, finding that different object classes and subclasses are separated despite being unlabelled. We demonstrate the interpretability of this latent space by identifying tracks within it that correspond to various spectral characteristics. For example, we find tracks that correspond to increasing star formation and increase in broad emission lines along the Balmer series. In upcoming work we hope to apply the methods presented here to search for both systematics and astrophysically interesting objects in much larger datasets of DESI spectra.

Nicolaou, C. [University Coll. London] (ORCID:0000↗

DOE OSTI2024

Assessing decision boundaries under uncertainty

In order to make design decisions, engineers may seek to identify regions of the design domain that are acceptable in a computationally efficient manner. A design is typically considered acceptable if its reliability with respect to parametric uncertainty exceeds the designer’s desired level of confidence. Despite major advancements in reliability estimation and in design classification via decision boundary estimation, the current literature still lacks a design classification strategy that incorporates parametric uncertainty and desired design confidence. To address this gap, this paper offers a novel interpretation of the acceptance region by defining the decision boundary as the hypersurface which isolates the designs that exceed a user-defined level of confidence given parametric uncertainty. This work addresses the construction of this novel decision boundary using computationally efficient algorithms that were developed for reliability analysis and decision boundary estimation. The approach proposed in this paper is verified on two physical examples from structural and thermal analysis using Support Vector Machines and Efficient Global Optimization-based contour estimation.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2025

Computationally efficient models for aqueous organic redox flow batteries

The rising usage of intermittent energy has garnered the need for large scale energy storage systems. Redox flow batteries (RFB) based energy storage system shows promising potential. Numerical simulations and machine learning approaches have been widely used to study RFB performance. The development of autonomous material discovery framework and digital twin of energy storage system usually needs to query cell performance through fast response models. In this study, two computationally efficient models are introduced: a physics-based analytical flow battery model (EZBattery), and a machine learning operator model (Deep Operator Network, denoted by DeepONet). Both models can provide cell performance near instantly, and prediction accuracy was systematically examined on an application of evaluating the performances of a 780 cm 2 aqueous organic redox flow battery (AORFB), using potential anolyte candidates in dihydroxyphenazine (DHP)-based family of organic materials. A validated computationally expansive 3-dimensional multi-physics finite element model by COMSOL was used as the ground truth and provided the training data set for the DeepONet. 1280 samples were generated with 10 properties to mimic the different possible anolyte candidates, and the cell performances were evaluated under 10 different combined operating conditions. The accuracy comparisons for the two computationally efficient models show that both models can provide comparable accuracy in predicting cell charging/discharging voltage curves. DeepONet can provide slightly higher overall accuracy than EZBattery with faster calculation speed, but highly relies on the training dataset. EZBattery does not need a training dataset and can provide interpretable physics-based explanations of the results, while being more flexible to adjust to adapt any different cell designs, flow battery architectures, and electrolyte materials.

Analytical model↗

DOE OSTI2025

Replacing non-biomedical concepts improves embedding of biomedical concepts

Embeddings are semantically meaningful representations of words in a vector space, commonly used to enhance downstream machine learning applications. Traditional biomedical embedding techniques often replace all synonymous words representing biological or medical concepts with a unique token, ensuring consistent representation and improving embedding quality. However, the potential impact of replacing non-biomedical concept synonyms has received less attention. Embedding approaches often employ concept replacement to replace concepts that span multiple words, such as non-small-cell lung carcinoma, with a single concept identifier (e.g., D002289). Also, all synonyms of each concept are merged into the same identifier. Here, we additionally leveraged WordNet to identify and replace sets of non-biomedical synonyms with their most common representatives. This combined approach aimed to reduce embedding noise from non-biomedical terms while preserving the integrity of biomedical concept representations. We applied this method to 1,055 biomedical concept sets representing molecular signatures or medical categories and assessed the mean pairwise distance of embeddings with and without non-biomedical synonym replacement. A smaller mean pairwise distance was interpreted as greater intra-cluster coherence and higher embedding quality. Embeddings were generated using the Word2Vec algorithm applied to a corpus of 10 million PubMed abstracts. Our results demonstrate that the addition of non-biomedical synonym replacement reduced the mean intra-cluster distance by an average of 8%, suggesting that this complementary approach enhances embedding quality. Future work will assess its applicability to other embedding techniques and downstream tasks. Python code implementing this method is provided under an open-source license.

algorithms↗

DOE OSTI2020

A machine learning approach to emulation and biophysical parameter estimation with the Community Land Model, version 5

Abstract. Land models are essential tools for understanding and predicting terrestrial processes and climate–carbon feedbacks in the Earth system, but uncertainties in their future projections are poorly understood. Improvements in physical process realism and the representation of human influence arguably make models more comparable to reality but also increase the degrees of freedom in model configuration, leading to increased parametric uncertainty in projections. In this work we design and implement a machine learning approach to globally calibrate a subset of the parameters of the Community Land Model, version 5 (CLM5) to observations of carbon and water fluxes. We focus on parameters controlling biophysical features such as surface energy balance, hydrology, and carbon uptake. We first use parameter sensitivity simulations and a combination of objective metrics including ranked global mean sensitivity to multiple output variables and non-overlapping spatial pattern responses between parameters to narrow the parameter space and determine a subset of important CLM5 biophysical parameters for further analysis. Using a perturbed parameter ensemble, we then train a series of artificial feed-forward neural networks to emulate CLM5 output given parameter values as input. We use annual mean globally aggregated spatial variability in carbon and water fluxes as our emulation and calibration targets. Validation and out-of-sample tests are used to assess the predictive skill of the networks, and we utilize permutation feature importance and partial dependence methods to better interpret the results. The trained networks are then used to estimate global optimal parameter values with greater computational efficiency than achieved by hand tuning efforts and increased spatial scale relative to previous studies optimizing at a single site. By developing this methodology, our framework can help quantify the contribution of parameter uncertainty to overall uncertainty in land model projections.

54 ENVIRONMENTAL SCIENCES↗

DOE OSTI2021

Optimal decision trees for categorical data via integer programming

Decision trees have been a very popular class of predictive models for decades due to their interpretability and good performance on categorical features. However, they are not always robust and tend to overfit the data. Additionally, if allowed to grow large, they lose interpretability. In this paper, we present a mixed integer programming formulation to construct optimal decision trees of a prespecified size. We take the special structure of categorical features into account and allow combinatorial decisions (based on subsets of values of features) at each node. Our approach can also handle numerical features via thresholding. Here we show that very good accuracy can be achieved with small trees using moderately-sized training sets. The optimization problems we solve are tractable with modern solvers.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2026

Remote sensing of planetary boundary layer from ground and space: Structure, thermodynamics, and boundary layer clouds

The planetary boundary layer (PBL) is a critical interface between Earth's surface and atmosphere, influencing atmospheric convection, weather patterns, and air quality. Recognized by the 2017 National Academies of Sciences, Engineering, and Medicine Earth Science Decadal Survey as an Incubation Targeted Observable, high-quality and effective observation of the PBL has become a priority. This paper reviews surface-based and satellite remote sensing techniques for characterizing PBL features, including PBL height (PBLH), boundary layer thermodynamics, turbulence, and boundary layer clouds (BLCs). These elements are treated as interconnected aspects of the PBL system, while recognizing that different instruments retrieve different physical manifestations of the PBL. The review summarizes recent advances and limitations in Micro-Pulse Lidar (MPL), Doppler lidar, Raman lidar, Differential Absorption Lidar (DIAL), ceilometers, wind profilers, GNSS Radio Occultation, radar, and hyperspectral sounders. These observations help trace dynamic processes within the PBL and link PBL structure to broader weather and climate processes. The review also highlights persistent observational gaps over oceans, remote land, and polar regions, where continuous surface-based profiling is sparse. By comparing instrument capabilities, practical limitations, and interpretation issues, this review emphasizes the need for integrated remote sensing approaches and careful definition of the retrieved PBL quantities.

Boundary layer clouds↗

DOE OSTI2023

Code for the manuscript titled "Blackout Diffusion: Generative Diffusion Models in Discrete-State Spaces"

We would like to disclose two scripts, written in Jupyter notebook, in which we implement the "Blackout Diffusion Process" described in the manuscript "Blackout Diffusion: Generative Diffusion Models in Discrete-State Spaces" (LA-UR-23-20509), to be submitted to the International Conference of Machine Learning (ICML). The abstract of the manuscript is append below. == Typical generative diffusion models rely on a Gaussian diffusion process for training the backward transformations, which can then be used generate samples from Gaussian noise. However, real world data often takes place in discrete-state spaces, which occur in many scientific applications. Here, we develop a theoretical formulation for arbitrary discrete-state Markov processes in the forward diffusion process. We relate the theory to the existing continuous-state Gaussian diffusion and identify the corresponding reverse-time stochastic process and score function in the continuous-time setting, and the reverse-time mapping in the discrete-time setting. As an example of this framework, we introduce "Blackout Diffusion", which learns to produce samples from an empty image instead of from noise. Numerical experiments on the CIFAR-10 dataset confirm the feasibility of generative diffusion modeling in a discrete space. Generalizing from specific (Gaussian) forward processes to a more general framework also sheds light on how to interpret generative diffusion models and their mathematical structure, which we comment on.

Lin, Yen Ting↗