Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “interpretable models”

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 559 records · Page 31
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Using Ultrasound Image Augmentation and Ensemble Predictions to Prevent Machine-Learning Model Overfitting

Deep learning predictive models have the potential to simplify and automate medical imaging diagnostics by lowering the skill threshold for image interpretation. However, this requires predictive models that are generalized to handle subject variability as seen clinically. Here, we highlight methods to improve test accuracy of an image classifier model for shrapnel identification using tissue phantom image sets. Using a previously developed image classifier neural network—termed ShrapML—blind test accuracy was less than 70% and was variable depending on the training/test data setup, as determined by a leave one subject out (LOSO) holdout methodology. Introduction of affine transformations for image augmentation or MixUp methodologies to generate additional training sets improved model performance and overall accuracy improved to 75%. Further improvements were made by aggregating predictions across five LOSO holdouts. This was done by bagging confidences or predictions from all LOSOs or the top-3 LOSO confidence models for each image prediction. Top-3 LOSO confidence bagging performed best, with test accuracy improved to greater than 85% accuracy for two different blind tissue phantoms. This was confirmed by gradient-weighted class activation mapping to highlight that the image classifier was tracking shrapnel in the image sets. Overall, data augmentation and ensemble prediction approaches were suitable for creating more generalized predictive models for ultrasound image analysis, a critical step for real-time diagnostic deployment.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2025

ACES-GNN: can graph neural network learn to explain activity cliffs?

Graph Neural Networks (GNNs) have revolutionized molecular property prediction by leveraging graph-based representations, yet their opaque decision-making processes hinder broader adoption in drug discovery. This study introduces the Activity-Cliff-Explanation-Supervised GNN (ACES-GNN) framework, designed to simultaneously improve predictive accuracy and interpretability by integrating explanation supervision for activity cliffs (ACs) into GNN training. ACs, defined by structurally similar molecules with significant potency differences, pose challenges for traditional models due to their reliance on shared structural features. By aligning model attributions with chemist-friendly interpretations, the ACES-GNN framework bridges the gap between prediction and explanation. Validated across 30 pharmacological targets, ACES-GNN consistently enhances both predictive accuracy and attribution quality for ACs compared to unsupervised GNNs. Our results demonstrate a positive correlation between improved predictions and accurate explanations, offering a robust and adaptable framework to better understand and interpret ACs. This work underscores the potential of explanation-guided learning to advance interpretable artificial intelligence in molecular modeling and drug discovery.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2023

High-throughput Li plating quantification for fast-charging battery design

Fast charging of most commercial lithium-ion batteries is limited due to fear of lithium plating on the graphite anode, which is difficult to detect and poses considerable safety risk. In this report we demonstrate the power of simple, accessible and high-throughput cycling techniques to quantify irreversible Li plating spanning data from over 200 cells. We first observe the effects of energy density, charge rate, temperature and state of charge on lithium plating, use the results to refine a mature physics-based electrochemical model and provide an interpretable empirical equation for predicting the plating onset state of charge. We then explore the reversibility of lithium plating and its connection to electrolyte design for preventing irreversible Li accumulation. Finally, we design a method to quantify in situ Li plating for commercially relevant graphite|LiNi 0.5 Mn 0.3 Co 0.2 O 2 (NMC) cells and compare with results from the experimentally convenient Li|graphite configuration. The hypotheses and abundant data herein were generated primarily with equipment universal to the battery researcher, encouraging further development of innovative testing methods and data processing that enable rapid battery engineering.

25 ENERGY STORAGE↗
DOE OSTI2021

Feedback density and causal complexity of simulation model structure

Measures of simulation model complexity generally focus on outputs; we propose measuring the complexity of a model’s causal structure to gain insight into its fundamental character. This article introduces tools for measuring causal complexity. First, we introduce a method for developing a model’s causal structure diagram, which characterises the causal interactions present in the code. Causal structure diagrams facilitate comparison of simulation models, including those from different paradigms. Next, we develop metrics for evaluating a model’s causal complexity using its causal structure diagram. We discuss cyclomatic complexity as a measure of the intricacy of causal structure and introduce two new metrics that incorporate the concept of feedback, a fundamental component of causal structure. The first new metric introduced here is feedback density, a measure of the cycle-based interconnectedness of causal structure. The second metric combines cyclomatic complexity and feedback density into a comprehensive causal complexity measure. Finally, we demonstrate these complexity metrics on simulation models from multiple paradigms and discuss potential uses and interpretations. These tools enable direct comparison of models across paradigms and provide a mechanism for measuring and discussing complexity based on a model’s fundamental assumptions and design.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Training custom light curve models of SN Ia subpopulations selected according to host galaxy properties

ABSTRACT Type Ia supernova (SN Ia) cosmology analyses include a luminosity step function in their distance standardization process to account for an observed yet unexplained difference in the post-standardization luminosities of SNe Ia originating from different host galaxy populations [e.g. high-mass ($M \gtrsim 10^{10} \, {\rm M}_{\odot }$) versus low-mass galaxies]. We present a novel method for including host-mass correlations in the SALT3 (Spectral Adaptive Light curve Template 3) light curve model used for standardizing SN Ia distances. We split the SALT3 training sample according to host-mass, training independent models for the low- and high-host-mass samples. Our models indicate that there are different average Si ii spectral feature strengths between the two populations, and that the average spectral energy distribution of SNe from low-mass galaxies is bluer than the high-mass counterpart. We then use our trained models to perform an SN cosmology analysis on the 3-yr spectroscopically confirmed Dark Energy Survey SN sample, treating SNe from low- and high-mass host galaxies as separate populations throughout. We find that our mass-split models reduce the Hubble residual scatter in the sample, albeit at a low statistical significance. We do find a reduction in the mass-correlated luminosity step but conclude that this arises from the model-dependent re-definition of the fiducial SN absolute magnitude rather than the models themselves. Our results stress the importance of adopting a standard definition of the SN parameters (x0, x1, c) in order to extract the most value out of the light curve modelling tools that are currently available and to correctly interpret results that are fit with different models.

Taylor, G. (ORCID:0000000157563259)↗
DOE OSTI2020

Using Fourier-Plane Imaging Microscopy for Determining Transition-Dipole-Moment Orientations in Organic Light-Emitting Devices

Here, we use Fourier-plane imaging microscopy (FIM) to determine the transition-dipole-moment orientation in doped organic emissive thin films. The use of FIM enables precise, sensitive, and rapid measurement of dipole orientation in the emission layer of an organic light-emitting device (OLED). An optical model of a stratified birefringent multilayer is introduced for interpreting results obtained by FIM. Using the model, we determine the average orientation of transition-dipole-moment vectors of three phosphorescent dopant emitters. The dipole alignment measured by FIM quantitatively explains the difference in OLED efficiencies using these archetype dopant molecules. FIM provides a nondestructive tool to measure and ultimately improve the outcoupling efficiency of OLEDs and other light-emitting devices.

74 ATOMIC AND MOLECULAR PHYSICS↗
DOE OSTI2024

PDF DECODER ANALYSIS CODE

SF-24-038"PDFdecoder", as a new application to explore parametrizations of parton distribution functions (PDFs) of the proton or other hadrons. The PDFs are fundamental quantities in particle physics which are necessary inputs to precise theoretical predictions for experiments at the Large Hadron Collider (LHC) and other facilities. As such, understanding how the PDFs are parametrized and associated uncertainties is a pressing need. The specific problem PDFdecoder confronts is the need of having a tractable and interpretably machine-learning (ML) framework to parametrize the PDFs and their uncertainties so as to understand how a given preferred parametrization is obtained. This problem has not been significantly addressed in the current literature. While other groups have used ML-based approaches to parametrize PDFs in the form of feed-forward neural networks, the question of tractability has not been explored in a PDF context. Our solution makes significant progress in this problem by using an array of encoder-decoder (essentially, autoencoder) architectures with varying constraints to the intermediate latent spaces based on interpretable physics. As a consequence, the trained models can be used as generative networks to produce interpretable predictions for the PDFs in a way that can be refined and studied further.

Hobbs, Timothy↗
DOE OSTI2020

Composition of Magma and Characteristics of the Hydrothermal System of Newberry Volcano, Oregon, From Magnetotellurics

Abstract We use 3‐D magnetotellurics to improve our understanding of the structure and magma composition of Newberry Volcano in Oregon, USA. Newberry is a broad shield volcano with a summit caldera and strongly bimodal magmatism. Newberry has long been the subject of geothermal exploration research, but that work has focused on the volcano's west flank, leaving the caldera largely unstudied with geophysical methods until recently. Our modeling shows a relatively resistive magma reservoir of approximately 50 Ωm. Our work builds upon recent seismic models and petrological analysis to interpret Newberry's magma reservoir as a dry rhyolite with 8–11% partial melt, which matches the seismically determined melt fraction. Finding the conditions within the magma reservoir that allow the resistivity and seismic analysis to come to the same melt fraction helps us narrow down the magma temperature and composition. From this we infer a dry rhyolitic magma at 850 °C. We also image a prominent vertical conductive anomaly along the south rim below the vent that produced the most recent eruption. The anomaly extends from magma reservoir depths of 3 km to 1 km below the surface where it disperses into the caldera fill. We interpret this as the main conduit for magmatic fluids to reach Newberry's hydrothermal system. Other features that our model shows include higher conductivity along the caldera rim and a resistive older pluton on the west flank.

Bowles‐Martinez, E.↗
DOE OSTI2023

Making Corgis Important for Honeycomb Classification: Adversarial Attacks on Concept-based Explainability Tools

Methods for model explainability have become increasingly critical for testing the fairness and soundness of deep learning. Concept-based interpretability techniques, which use a small set of human-interpretable concept exemplars in order to measure the influence of a concept on a model's internal representation of input, are an important thread in this line of research. In this work we show that these explainability methods can suffer the same vulnerability to adversarial attacks as the models they are meant to analyze. We demonstrate this phenomenon on two well-known concept-based interpretability methods: TCAV and faceted feature visualization. We show that by leveraging the geometry of the problem and carefully perturbing the examples of the concept that is being investigated, we can radically change the output of the interpretability method. The attacks that we propose can either induce positive interpretations (polka dots are an important concept for a model when classifying zebras) or negative interpretations (stripes are not an important factor in identifying images of a zebra). Our work highlights the fact that in safety-critical applications, there is need for security around not only the machine learning pipeline but also the model interpretation process.

Brown, Davis R.↗
DOE OSTI2025

Hydrogen adsorption and transport in clay-rich geomaterials: Implications for large-volume underground hydrogen storage

Depleted oil and gas reservoirs, characterized by impermeable clay-rich caprocks, are promising sites for large-scale underground hydrogen storage (UHS), which is a key strategy to support hydrogen-based energy systems. However, experimental data on hydrogen storage in clay-rich geomaterials remain scarce. In this work, we experimentally investigated hydrogen adsorption and migration in clay-rich geomaterials in the presence of nitrogen and water under controlled temperatures. Experimental observations showed that hydrogen was adsorbed in dry illite. A dual-porosity transport model was developed to interpret hydrogen transport between large-pore and small-pore domains in illite. The large-pore domain is the space between clay particles (i.e., inter-particle space), whereas the small-pore domain is the nanoscale pore space between clay mineral layers (i.e., inter-layer or intra-particle space). In contrast, nitrogen showed no evidence of adsorption in dry illite because it cannot move into the small-pore domain due to the relatively large kinetic diameter, referred to as the molecular sieving effect. Here, we found that 0.7–1.3 nm is the length scale regulating this molecular sieving effect, matching the interlayer spacing in illite, suggesting that nitrogen is a promising cushion gas in UHS, which aims to maintain adequate pressure in the reservoir for economic operations. In wetted illite, hydrogen was not adsorbed into the interlayer space due to the occupation of adsorption sites by interlayer water, which highlights the critical role of the clay hydration state in controlling hydrogen-clay interactions. Additionally, hydrogen adsorption experiments on crushed shale indicated that the shale surface possessed adsorption sites more favorable for hydrogen than for nitrogen. Through these experiments, we provide new insights into hydrogen storage mechanisms in clay-rich geomaterials and offer valuable laboratory data for evaluating the performance of large-scale UHS systems.

Adsorption↗
DOE OSTI2026

Generative models on phase space

Deep generative models such as diffusion and flow matching are powerful machine learning tools capable of learning and sampling from high-dimensional distributions. They are particularly useful when the training data appears to be concentrated on a submanifold of the data embedding space. For high-energy physics data, consisting of collections of relativistic energy-momentum 4-vectors, this submanifold can enforce extremely strong physically-motivated priors, such as energy and momentum conservation. If these constraints are learned only approximately, rather than exactly, this can inhibit the interpretability and reliability of such generative models. To remedy this deficiency, we introduce generative models which are, by construction, confined at every step of their sampling trajectory to the manifold of massless N-particle Lorentz-invariant phase space in the center-of-momentum frame. In the case of diffusion models, the "pure noise" forward process endpoint corresponds to the uniform distribution on phase space, which provides a clear starting point from which to identify how correlations among the particles emerge during the reverse (de-noising) process. We demonstrate that our models are able to learn both few-particle and many-particle distributions with various singularity structures, paving the way for future interpretability studies using generative models trained on simulated jet data.

Bogorad, Zachary [Fermilab]↗
DOE OSTI2021

Protein identification from electron cryomicroscopy maps by automated model building and side-chain matching

Using single-particle electron cryo-microscopy (cryo-EM), it is possible to obtain multiple reconstructions showing the 3D structures of proteins imaged as a mixture. Here, it is shown that automatic map interpretation based on such reconstructions can be used to create atomic models of proteins as well as to match the proteins to the correct sequences and thereby to identify them. This procedure was tested using two proteins previously identified from a mixture at resolutions of 3.2 Å, as well as using 91 deposited maps with resolutions between 2 and 4.5 Å. The approach is found to be highly effective for maps obtained at resolutions of 3.5 Å and better, and to have some utility at resolutions as low as 4 Å.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2020

Magnetic field frustration of the metal-insulator transition in V 2 O 3

Despite decades of efforts, the origin of metal-insulator transitions (MITs) in strongly correlated materials remains one of the main long-standing problems in condensed-matter physics. An archetypal example is V 2 O 3 , which undergoes simultaneous electronic, structural, and magnetic phase transitions. This remarkable feature highlights the many degrees of freedom at play in this material. In this work, acting solely on the magnetic degree of freedom, we reveal an anomalous feature in the electronic transport of V 2 O 3 : On cooling, the magnetoresistance changes from positive to negative values well above the MIT temperature, and shows divergent behavior at the transition. In this work, the effects are attributed to the magnetic field quenching antiferromagnetic fluctuations above the Néel temperature $T_N$, and preventing long-range antiferromagnetic ordering below $T_N$. In both cases, suppressing the antiferromagnetic ordering prevents the opening of the incipient electronic gap. This interpretation is supported by Hubbard model calculations which fully reproduce the experimental behavior. Our study sheds light on this classic problem providing a clear and physical interpretation of the nature of the metal-insulator transition.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2020

Interpreting and Stabilizing Machine-Learning Parametrizations of Convection

Neural networks are a promising technique for parameterizing subgrid-scale physics (e.g., moist atmospheric convection) in coarse-resolution climate models, but their lack of interpretability and reliability prevents widespread adoption. For instance, it is not fully understood why neural network parameterizations often cause dramatic instability when coupled to atmospheric fluid dynamics. This paper introduces tools for interpreting their behavior that are customized to the parameterization task. First, we assess the nonlinear sensitivity of a neural network to lower-tropospheric stability and the midtropospheric moisture, two widely studied controls of moist convection. Second, we couple the linearized response functions of these neural networks to simplified gravity wave dynamics, and analytically diagnose the corresponding phase speeds, growth rates, wavelengths, and spatial structures. To demonstrate their versatility, these techniques are tested on two sets of neural networks, one trained with a super-parameterized version of the Community Atmosphere Model (SPCAM) and the second with a near-global cloud-resolving model (GCRM). Additionally, even though the SPCAM simulation has a warmer climate than the cloud-resolving model, both neural networks predict stronger heating/drying in moist and unstable environments, which is consistent with observations. Moreover, the spectral analysis can predict that instability occurs when GCMs are coupled to networks that support gravity waves that are unstable and have phase speeds larger than 5 m s -1 . In contrast, standing unstable modes do not cause catastrophic instability. Using these tools, differences between the SPCAM-trained versus GCRM-trained neural networks are analyzed, and strategies to incrementally improve both of their coupled online performance unveiled.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

Universal Stripe Symmetry of Short-Range Charge Density Waves in Cuprate Superconductors

The omnipresence of charge density waves (CDWs) across almost all cuprate families underpins a common organizing principle. However, a longstanding debate of whether its spatial symmetry is stripe or checkerboard remains unresolved. While CDWs in lanthanum- and yttrium-based cuprates possess a stripe symmetry, distinguishing these two scenarios is challenging for the short-range CDW in bismuth-based cuprates. Here, high-resolution resonant inelastic x-ray scattering is employed to uncover the spatial symmetry of the CDW in Bi 2 Sr 2-x La x CuO 6 + δ . Across a wide range of doping and temperature, anisotropic CDW peaks with elliptical shapes are found in reciprocal space. Based on Fourier transform analysis of real-space models, the results are interpreted as evidence of unidirectional charge stripes, hosted by mutually 90°-rotated anisotropic domains. This work paves the way for a unified symmetry and microscopic description of CDW order in cuprates.

36 MATERIALS SCIENCE↗
DOE OSTI2023

Search for new phenomena in final states with photons, jets and missing transverse momentum in $pp$ collisions at $\sqrt{s}$ = 13 TeV with the ATLAS detector

A search for new phenomena has been performed in final states with at least one isolated high-momentum photon, jets and missing transverse momentum in proton– proton collisions at a centre-of-mass energy of $\sqrt{s}$ = 13 TeV. The data, collected by the ATLAS experiment at the CERN LHC, correspond to an integrated luminosity of 139 fb –1 . The experimental results are interpreted in a supersymmetric model in which pair-produced gluinos decay into neutralinos, which in turn decay into a gravitino, at least one photon, and jets. No significant deviations from the predictions of the Standard Model are observed. Upper limits are set on the visible cross section due to physics beyond the Standard Model, and lower limits are set on the masses of the gluinos and neutralinos, all at 95% confidence level. Visible cross sections greater than 0.022 fb are excluded and pair-produced gluinos with masses up to 2200 GeV are excluded for most of the NLSP masses investigated.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2023

Search for long-lived particles using out-of-time trackless jets in proton-proton collisions at $\sqrt{s}$ = 13 TeV

A search for long-lived particles decaying in the outer regions of the CMS silicon tracker or in the calorimeters is presented. The search is based on a data sample of proton-proton collisions at $\sqrt{s}$ = 13 TeV recorded with the CMS detector at the LHC in 2016–2018, corresponding to an integrated luminosity of 138 fb −1 . A novel technique, using nearly trackless and out-of-time jet information combined in a deep neural network discriminator, is employed to identify decays of long-lived particles. The results are interpreted in a simplified model of chargino-neutralino production, where the neutralino is the next-to-lightest supersymmetric particle, is long-lived, and decays to a gravitino and either a Higgs or Z boson. This search is most sensitive to neutralino proper decay lengths of approximately 0.5 m, for which masses up to 1.18 TeV are excluded at 95% confidence level. The current search is the best result to date in the mass range from the kinematic limit imposed by the Higgs boson mass up to 1.8 TeV.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2020

Local quenches, bulk entanglement entropy and a unitary Page curve

Quantum corrections to the entanglement entropy of matter fields interacting with dynamical gravity have proven to be very important in the study of the black hole information problem. We consider a one-particle excited state of a massive scalar field infalling in a pure AdS 3 geometry and compute these corrections for bulk subregions anchored on the AdS boundary. In the dual CFT 2 , the state is given by the insertion of a local primary operator and its evolution thereafter. We calculate the area and bulk entanglement entropy corrections at order $$ \mathcal{O}\left({N}^0\right), $$ O N 0 , both in AdS and its CFT dual. The two calculations match, thus providing a non-trivial check of the FLM formula in a dynamical setting. Further, we observe that the bulk entanglement entropy follows a Page curve. We explain the precise sense in which our setup can be interpreted as a simple model of black hole evaporation and comment on the implications for the information problem.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗