Towards the Development of a Framework for Interpretable Hierarchical Calibration of Agent-Based Models
Explore the source record for details and available documents.
Acquesta, Erin Carolyn Solfiell↗
SEARCH · Search NASA
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.
Neural networks for estimation of divertor conditions in DIII-D using C III imaging
Deep learning approaches have been applied to images of C III emission in the lower divertor of DIII-D to develop models for estimating the level of detachment and magnetic configuration (X-point location and strike point radial location). The poloidal distance from the target to the C III emission front is used to represent the level of detachment. The models perform well on a test dataset not used in training, achieving $F_1$ scores as high as 0.99 for detachment state classification and root mean squared error (RMSE) as low as 2cm for front location regression. Predictions for shots with intermittent reattachment are studied, with class activation mapping used to aid in interpretation of the model predictions. Based on the success of these models, a third model was trained to predict the X-point location and strike point radial position from C III images. Though the dataset covers only a small range of possible magnetic configurations, the model shows promising results, achieving RMSE around 1cm for the test data.
70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
Scientific machine learning for closure models in multiscale problems: A review
Here, closure problems are omnipresent when simulating multiscale systems, where some quantities and processes cannot be fully prescribed despite their effects on the simulation's accuracy. Recently, scientific machine learning approaches have been proposed as a way to tackle the closure problem, combining traditional (physics-based) modeling with data-driven (machine-learned) techniques, typically through enriching differential equations with neural networks. This paper reviews the different reduced model forms, distinguished by the degree to which they include known physics, and the different objectives of a priori and a posteriori learning. The importance of adhering to physical laws (such as symmetries and conservation laws) in choosing the reduced model form and choosing the learning method is discussed. The effect of spatial and temporal discretization and recent trends toward discretization-invariant models are reviewed. In addition, we make the connections between closure problems and several other research disciplines: inverse problems, Mori-Zwanzig theory, and multi-fidelity methods. In conclusion, much progress has been made with scientific machine learning approaches for solving closure problems, but many challenges remain. In particular, the generalizability and interpretability of learned models is a major issue that needs to be addressed further.
97 MATHEMATICS AND COMPUTING↗
Multi-Omics Reveals Temporal Scales of Carbon Metabolism in Synechococcus Elongatus PCC 7942 Under Light Disturbance
Central carbon metabolism in model cyanobacteria involves multiple pathways to adapt to energy-light limitations across diel cycles. However, the success in mechanistic modeling for phenotypic prediction of the protein regulators in the metabolic state depends on capturing the vast possibilities emerging from multiple regulatory pathways in complex biological processes. Here, we developed a physics-informed machine learning approach based on energy-landscape concepts to predict regulatory proteins responding to cyclic circadian and unforeseen light perturbations in cyanobacterial metabolic networks. Our approach provides interpretable de novo models for inferring gene expression dynamics from Synechococcus elongatus over diel cycles and using redox proteome analysis to distinguish immediate light-responsive elements from circadian-regulated processes in carbon metabolism pathways. We identified distinct temporal signatures with the analysis of the redox proteome: there was an immediate shift in cysteine redox states accompanied by a limited change in protein abundance under constant illumination and after 2 hours of darkness. This discovery indicates that the generation of reductants coordinates photoinduced electron transport with redox metabolic pathways in two discernable molecular mechanisms: fast redox-based protein modifications occur immediately after the light disturbance, followed by slow transcriptional regulations across networks. This temporal regulation reveals how metabolic networks integrate rapid light responses with programmed circadian rhythms to maintain cellular homeostasis under the light-energy limitations over the diel cycle.
Biomolecular & subcellular processes↗
Interpretation of Ion Irradiation and Neutron Irradiation Damage in Additively Manufactured 316 Stainless Steel using Multiscale Modeling
The accelerated adoption of nuclear energy necessitates advanced manufacturing technologies, such as additive manufacturing, to meet heightened supply chain requirements and support innovative reactor technologies. Due to the unique microstructural characteristics of additively manufactured materials under distinct solidification conditions, comprehensive evaluation of their performance in reactor environments is essential. The Advanced Materials and Manufacturing Technologies program under the Department of Energy's Office of Nuclear Energy focuses on understanding the irradiation performance and damage evolution of laser powder bed fusion 316 stainless steel, with an emphasis on integrating ion and neutron irradiation data to accelerate the development and qualification of materials for advanced nuclear reactor applications. While ion irradiation is a cost- and time-effective method, modeling and simulation are required to interpret the data for the broader range of irradiation conditions encountered in advanced reactors. In fiscal year 2025, integrated multiscale modeling and simulations were conducted to assess irradiation damage in additively manufactured 316 stainless steel. Key outcomes include predictions of chromium enrichment at grain boundaries, nickel enrichment at dislocation cell walls and void surfaces, and heterogeneous void evolution under ion and neutron irradiation conditions. Cluster dynamics simulations revealed the coarsening of voids at high irradiation temperatures and the suppression of void growth by high network dislocation density, while also demonstrating significant growth and coarsening of voids and self-interstitial atom loops at low dose rates. Machine learning-accelerated atomistic simulations highlighted the impact of the local environment and chromium concentration on vacancy diffusivity, providing key insights on the influence of composition on void swelling and radiation-induced segregation. Additionally, molecular dynamics simulations demonstrated the presence of defect production bias and a significant effect of carbon content on defect cluster behavior. These combined efforts aim to predict the performance of additively manufactured materials under various reactor conditions, supporting their qualification for nuclear reactor applications by interpreting ion irradiation data. This report underscores the potential of integrated multiscale modeling to analyze ion irradiation data in the effort to accelerate the qualification of additively manufactured materials for nuclear reactor components.
316 stainless steel↗
Charge Radii of Neutron-Rich Scandium Isotopes and the Seniority Symmetry in the 0𝑓 7/2 Shell
Nuclear charge radii of neutron-rich 47–49 Sc isotopes were measured using collinear laser spectroscopy at CERN-ISOLDE. The new data reveal that the charge radii of scandium isotopes exhibit a distinct trend between 𝑁 = 20 and 𝑁 = 28, with 41 Sc and 49 Sc isotopes having similar values, mirroring the closeness of the charge radii of 40 Ca and 48 Ca. Theoretical models that successfully interpret the radii of calcium isotopes could not account for the observed behavior in scandium radii, in particular the reduced odd-even staggering. Remarkably, the inclusion of the new 49 Sc radius data has unveiled a similar trend in the charge radii of 𝑁 = 28 isotones and 𝑍 = 20 isotopes when adding the protons atop the 48 Ca core and the neutrons atop the 40 Ca core, respectively. We demonstrate that this trend is consistent with the prediction of the seniority model.
Nuclear charge distribution↗
Sim-to-real supervised domain adaptation for radioisotope identification
Machine learning has the potential to improve the speed and reliability of radioisotope identification using gamma spectroscopy. However, meticulously labeling an experimental dataset for training is often prohibitively expensive, while training models purely on synthetic data is risky due to the domain gap between simulated and experimental measurements. In this research, we demonstrate that supervised domain adaptation can substantially improve the performance of radioisotope identification models by transferring knowledge between synthetic and experimental data domains. We consider two domain adaptation scenarios: (1) a simulation-to-simulation adaptation, where we perform multi-label proportion estimation using simulated high-purity germanium detectors, and (2) a simulation-to-experimental adaptation, where we perform multi-class, single-label classification using measured spectra from handheld lanthanum bromide (LaBr) and sodium iodide (NaI) detectors. We begin by pretraining a spectral classifier on synthetic data using a custom transformer-based neural network. After subsequent fine-tuning on just 64 labeled experimental spectra, we achieve a test accuracy of 96% in the sim-to-real scenario with a LaBr detector, far surpassing a synthetic-only baseline model (75%) and a model trained from scratch (80%) on the same 64 spectra. Furthermore, we demonstrate that domain-adapted models learn more human-interpretable features than experiment-only baseline models. Overall, our results highlight the potential for supervised domain adaptation techniques to bridge the sim-to-real gap in radioisotope identification, enabling the development of accurate and explainable classifiers even in real-world scenarios where access to experimental data is limited.
Lalor, Peter W.↗
Decoding substrate specificity determining factors in glycosyltransferase-B enzymes – insights from machine learning models
Substrate specificity is an essential characteristic of any enzyme's function and an understanding of the factors that determine this specificity is crucial for enzyme engineering. Unlike the structure of an enzyme which is directly impacted by its sequence, substrate specificity as an enzyme attribute involves a rather indirect relationship with sequence as it also depends on structural aspects that dictate substrate accessibility and active site dynamics. In this study, we explore the performance of classifier-based machine learning models trained on curated sequence and structural data for a class of glycosyltransferases (GTs), namely GT-Bs, to understand their substrate specificity determining factors. GTs enable the transfer of sugar moieties to other biomolecules such as oligosaccharides or proteins and are found in all kingdoms of life. In plants, GTs participate in the biosynthesis of plant cell wall biopolymers (e.g.: hemicelluloses and pectins) and are an integral part of the enzymatic machinery that enables the storage of carbon and energy as plant biomass. To elucidate the substrate specificity of uncharacterized GT-Bs, we constructed multi-label machine learning models (Support Vector Classifier, K-Nearest Neighbors, Gaussian Naïve-Bayes, Random Forest) that incorporate both sequence and structural features. These models achieve good predictive accuracies on test datasets. However, despite our use of structural information, we highlight that there is further scope for improvement in training these models to draw interpretable relationships between sequence, structure and substrate specificity determining motifs in GT-Bs.
97 MATHEMATICS AND COMPUTING↗
Attention-based explainability for structure–property relationships
Machine learning methods are emerging as a universal paradigm for constructing correlative structure–property relationships in materials science based on multimodal characterization. However, this necessitates the development of methods for the physical interpretability of the resulting correlative models. Here, we demonstrate the potential of attention-based neural networks for revealing structure–property relationships and the underlying physical mechanisms, using the ferroelectric properties of PbTiO3 thin films as a case study. Through the analysis of attention scores, we disentangle the influence of distinct domain patterns on the polarization switching process. The attention-based Transformer model is explored both as a direct interpretability tool and as a surrogate for explaining representations learned via unsupervised machine learning, enabling the identification of physically grounded correlations. We compare attention-derived interpretability scores with classical SHapley Additive exPlanations analysis and show that, in contrast to applications in natural language processing, attention mechanisms in materials science exhibit high efficiency in highlighting meaningful structural features.
Slautin, Boris [Independent Researcher]↗
A mixture parameterized biologically based dosimetry model to predict body burdens of polycyclic aromatic hydrocarbons in developmental zebrafish toxicity assays
Polycyclic aromatic hydrocarbons (PAHs) are a group of environmental toxicants found ubiquitously as complex mixtures in human-impacted environments. Developmental zebrafish exposures have been used widely to study PAH toxicity, but most studies report nominal exposure concentrations. Nominal exposure concentrations can be unreliable dose metrics due to differences in toxicant bioavailability resulting from disparate exposure methodologies and chemical properties. Toxicokinetic modeling can predict toxicant tissue doses to facilitate comparison between exposures of different chemicals, methodologies, and biological models. We parameterize a biologically based dosimetry model for developmental zebrafish toxicity assays for 9 PAHs. The model was optimized with measurements from media, tissue, and plastic plate walls throughout a static developmental exposure to a mixture of 10 PAHs of high abundance within the Portland Harbor Superfund Site. Plate binding, volatilization, zebrafish permeability, and tissue—media partitioning coefficients vary widely between PAHs. Model predictions accounted for 83% and 54% of 48 hpf body burdens within a factor of 2 resulting from exposures to mixtures and individual PAHs, respectively. Accounting for solubility significantly improves model performance. Competition for active sites in metabolizing enzymes may change biotransformation kinetics between individual PAH and mixture exposures. Area under the curve estimations of concentrations in zebrafish resulted in altered hazard rankings from nominal exposure concentrations. Future work will be oriented to generalizing the model to other PAHs. This PAH dosimetry model improves the interpretability of developmental zebrafish toxicity assays by providing time-resolved body burdens from nominal exposure concentrations.
63 RADIATION, THERMAL, AND OTHER ENVIRON. POLLUTAN↗
Search for a new resonance decaying into two spin-0 bosons in a final state with two photons and two bottom quarks in proton-proton collisions at $\sqrt{s}$ = 13 TeV
A search for a new boson X is presented using CERN LHC proton-proton collision data collected by the CMS experiment at $\sqrt{s}$ = 13 TeV in 2016–2018, and corresponding to an integrated luminosity of 138 fb -1 . The resonance X decays into either a pair of Higgs bosons HH of mass 125 GeV or an H and a new spin-0 boson Y. One H subsequently decays to a pair of photons, and the second H or Y, to a pair of bottom quarks. The explored mass ranges of X are 260–1000 GeV and 300–1000 GeV, for decays to HH and to HY, respectively, with the Y mass range being 90–800 GeV. For a spin-0 X hypothesis, the 95% confidence level upper limit on the product of its production cross section and decay branching fraction is observed to be within 0.90–0.04 fb, depending on the masses of X and Y. The largest deviation from the background-only hypothesis with a local (global) significance of 3.8 (below 2.8) standard deviations is observed for X and Y masses of 650 and 90 GeV, respectively. The limits are interpreted using several models of new physics.
72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
Establishing nationwide power system vulnerability index across US counties using interpretable machine learning
Power outages have become increasingly frequent, intense, and prolonged in the US due to climate change, aging electrical grids, and rising energy demand. However, largely due to the absence of granular spatiotemporal outage data, we lack data-driven evidence and analytics-based metrics to quantify power system vulnerability. This limitation has hindered the ability to effectively evaluate and address vulnerability to power outages in US communities. Here, in this work, we collected ∼179 million power outage records at 15-min intervals across 3022 US contiguous counties (96.15 % of the area) from 2014 to 2023. We developed a power system vulnerability assessment framework based on three dimensions (intensity, frequency, and duration) and applied interpretable machine learning models (XGBoost and SHAP) to compute Power System Vulnerability Index (PSVI) at the county level. Our analysis reveals a consistent increase in power system vulnerability across the US counties over the past decade. We identified 318 counties across 45 states as hotspots for high power system vulnerability, particularly in the West Coast (California and Washington), the East Coast (Florida and the Northeast area), the Great Lakes megalopolis (Chicago-Detroit metropolitan areas), and the Gulf of Mexico (Texas). Our heterogeneity analysis indicates that urban counties and those located along regional transmission boundaries tend to exhibit significantly higher vulnerability. Our results highlight the significance of the proposed PSVI for evaluating the vulnerability of communities to power outages. The findings underscore the widespread and pervasive impact of power outages across the country and offer crucial insights to support infrastructure operators, policymakers, and emergency managers in formulating policies and programs aimed at enhancing the resilience of the US power infrastructure.
24 POWER TRANSMISSION AND DISTRIBUTION↗
Structural characterization of high-protein system through ultra-small and small-angle X-ray scattering
High-protein systems exhibit a hierarchical structure consisting of interconnected multiscale length assemblies. Recent investigations are directed toward identifying these structural units under various processing and environmental conditions to establish structure–function correlations. Ultra-small and small-angle X-ray scattering (USAXS/SAXS) has become a crucial tool for characterizing the structures of proteins and their clusters/aggregates, ranging from nanometers to micrometers, with minimal disruption to their original state. Here, this review first describes the facilities and principles of X-ray scattering, followed by discussions on the analysis of scattering data, including the interpretation of fitting models. It then delves into the main applications of USAXS/SAXS in plant and dairy protein-rich systems. Future strategies to enhance the utilization of scattering techniques for elucidating the structure of high-protein systems are also included.
60 APPLIED LIFE SCIENCES↗
Control Mechanisms for Self‐Sealing in Activated Clay‐Rich Faults Through Controlled Hydraulic Injection Experiment
Abstract In a high‐pressure injection fault activation experiment conducted at the Mont Terri underground research laboratory in Switzerland, the transmissivity of the Opalinus Clay fault significantly increased due to opening and shearing. The fluid injection, spanning a few hours, generated a 10 m radius fault activation patch. Subsequent pressure pulse tests conducted bi‐weekly for a year revealed the gradual return of fault transmissivity to its initial state. The study utilized fluid pressure decay analysis, optical fiber monitoring, continuous active source seismic measurements and borehole displacement sensors for measuring fault displacements. The fault zone exhibited a dilation of approximately 1.4 mm, associated with both normal and tangential movements during activation, resulting in a sudden transmissivity increase from 1 × 10 −12 to 3.2 × 10 −7 m 2 /s. Early post‐activation, transient compaction and the subsequent slow compaction were observed, transitioning to an extension regime. The pressure pulse tests demonstrated a rapid transmissivity drop by more than two orders of magnitude within the first 10 days, followed by a gradual and less pronounced decrease. Plastic shear and compaction dominated the transmissivity evolution until 70 days after injection ended, followed by a period where additional factors, such as clay mineral swelling, influenced the behavior. Extrapolation suggested a sealing process taking at least 50 years after the initial activation. Plain Language Summary A field‐scale fault activation experiment offers valuable insights into the elasto‐plastic processes governing the sealing of shale faults. The experiment reveals a rapid increase in the fault's transmissivity by approximately five orders of magnitude during activation. Subsequent observations show a gradual transmissivity decrease by about three orders of magnitude post‐activation, with slow long‐term plastic shear and compaction of the fault competing against secondary processes, notably clay mineral swelling. All conceptual models employed to interpret these field data converge on the estimation that the fault's return to its initial low transmissivity state would require a minimum of 50 years. Key Points High‐pressure injection fault activation experiment at the Mont Terri underground research laboratory Continuous transmissivity measurements record self‐sealing inside a clay‐rich fault zone Transmissivity undergoes a phase of domination by slow plastic compaction and shearing during the initial post‐activation period, with mineral swelling exerting its influence over the long term
Guglielmi, Yves↗
In-beam spectroscopy reveals competing nuclear shapes in the rare isotope 62 Cr
In recent decades, rareisotope facilities have enabled the study of short-lived, neutron-rich nuclei. Their measured properties indicate that shell structure changes in the regime of unbalanced neutron-to-proton ratios compared to that of stable nuclei. In the so-called islands of inversion in the nuclear chart – around the neutron-rich nuclei 32 Mg, 42 Si and 64 Cr, for example – the textbook shell-model picture predicts the presence of spherical shapes due to the respective magic neutron numbers of 20, 28, and 40 of these nuclei. However, nuclei in these regions turn out to be deformed in their ground states. Another hallmark of these islands is shape coexistence, where a nucleus assumes different shapes with excitation energy. Furthermore, we present evidence for this phenomenon from the observation of an excited 0 + state in 62 Cr, two neutrons away from the heart of the island of inversion around neutron number N = 40. We use large-scale shell-model calculations to interpret the results and report extrapolations for the doubly-magic nucleus 60 Ca.
73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
Measurement of Thomson-scattering spectra with continuous angular resolution (invited)
Here, a novel Thomson-scattering diagnostic with continuous angular resolution over a span of 120° was developed for the characterization of plasmas produced at the Omega Laser Facility. Spectrally resolving light scattered from electron plasma wave features as a function of emission angle provides a means to efficiently probe a large range of plasma frequencies and k vectors. Together, these spectra contain critical constraints on the plasma-physics models used to interpret the data and enable experimental measurements of the electron-velocity distribution function over several orders of magnitude without assumptions about its mathematical form. Major components of the instrument include (1) a reflective collection objective that gathers light over a range of 120° × 12°; (2) a spatial-filter image relay for measurement localization; (3) cylindrical optics for producing a line image of the collection aperture; (4) a transmission grating spectrometer; and (5) a time-gated, image-intensified camera. Thomson-scattered light collected from an ~(50 – μM) 3 volume of plasma is recorded with 0.8-nm spectral and 1° angular resolution. Initial experiments examined the properties of the electron-velocity distribution in gas-jet-produced plasmas in the presence of heating via inverse bremsstrahlung absorption.
47 OTHER INSTRUMENTATION↗
Continental-Scale Controls on Hyporheic Respiration Revealed by Knowledge-Guided Machine Learning
Hyporheic zone sediments regulate organic matter turnover and in-stream respiration, yet controls on sediment respiration remain poorly constrained across heterogeneous river networks, limiting prediction of stream metabolism and carbon processing at continental scales. Here, we integrate observations from ~90 river corridors across the United States in the WHONDRS consortium with a knowledge-guided machine learning (KGML) framework that couples thermodynamic rate theory with machine learning to identify dominant controls on hyporheic respiration. Diagnostic analyses show that organic matter concentration and thermodynamic favorability define an upper bound on respiration potential, whereas biological catalytic capacity and physical accessibility jointly govern realized respiration rates through interaction effects. To represent unmeasurable accessibility constraints, we use the mechanistic model as a scaffold for KGML, allowing machine learning to target residual structure not explained by process theory. This hybrid framework improves predictive skill relative to both the mechanistic model alone and fully data-driven models while preserving interpretability. These results indicate that variability in hyporheic respiration is largely mechanistically structured and demonstrate how integrating process theory with explainable AI enhances predictive performance while enabling scalable synthesis of river corridor observations.
Zheng, Jianqiu↗
Explainable physics-based constraints on reinforcement learning for accelerator optimization
We present a reinforcement learning (RL) framework for optimizing particle accelerator experiments that builds explainable physics-based constraints on agent behavior. The goal is to increase transparency and trust by letting users verify that the agent’s decision-making process incorporates suitable physics. Our algorithm uses a learnable surrogate function for physical observables, such as energy, and uses them to fine-tune how actions are chosen. This surrogate can be represented by a neural network or by an interpretable sparse dictionary model. We test our algorithm on a range of particle accelerator optimization environments designed to emulate the Continuous Electron Beam Accelerator Facility at Jefferson Lab. By examining the mathematical form of the learned constraint function, we are able to confirm the agent has learned to use the established physics of each environment. In addition, we find that the introduction of a physics-based surrogate enables our RL algorithms to reliably converge for difficult high-dimensional accelerator optimization environments.
explainability↗