Search NASA⌕ Search

SEARCH · Search NASA

Results for “model selection”

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 289 records · Page 16
All sourcesNASA NTRSDOE OSTI
DOE OSTI2025

Explicit Dynamic Impact Analysis of the Building 3525 Packages

This report describes the methodology used for the safety basis analysis of the Building (Bldg.) 3525 casks and presents preliminary results for a selected geometry. The specific problem analyzed is a free fall load case onto an unyielding target at various orientations. The various orientations were analyzed to ensure that the packages were subjected to the orientation that causes the most damage. The analysis approach, material modeling, finite element modeling, and results for the selected case are presented, along with a brief discussion of conservatism and uncertainties. The results for the analyzed load cases show that through selected administrative controls, no breach of the container or loss of shielding will occur in the package.

42 ENGINEERING↗
DOE OSTI2025

Continuum shock mixture models for Ni+Al multilayers: Individual layers and bulk equations of state

Continuum shock mixture models are reviewed and applied to determine the equations of state for five different compositions of Ni x Al y ⁠, as well as bulk Ni+Al reactive multilayers, by combining the fundamental property data for elemental nickel and aluminum. From the literature, we down-select and evaluate two analytical models for the mixture Hugoniot, i.e., the well-known method of kinetic energy averaging (KEA) and a recent model proposed by Jordan and Baer [J. Appl. Phys. 111, 083516 (2012)]. Fundamentally, the former method assumes pressure equilibrium, whereas the latter assumes a common particle velocity and mixture sound speed from compressible two-phase cavitating flows. Additionally, we construct thermodynamically complete equations of state by fitting Einstein oscillator series models for the specific heat at constant volume. Finally, the solid solution approximation is invoked for intermetallic compositions, which are not strictly physical mixtures. Overall, the KEA model provides a better fit to the available Ni x Al y and Ni+Al multilayer shock compression data; however, there are combinations of material properties where the performance of these two models is thought to be reversed. Moreover, the results of this work include the first analytical solution of Jordan–Baer that does not require numerical root finding, as well as proposed modifications to the Einstein oscillator series to incorporate some effects of local pressure–temperature equilibrium and reaction–diffusion. Future work is planned that will use these equations of state in mesoscale simulations to study shock-induced reaction in Ni+Al multilayers, and the intended application is illustrated with a brief 2D hydrocode example.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Prediction of Dielectric Constant in Series of Polymers by Quantitative Structure-Property Relationship (QSPR)

This work is devoted to the investigation of dielectric permittivity which is influenced by electronic, ionic, and dipolar polarization mechanisms, contributing to the material’s capacity to store electrical energy. In this study, an extended dataset of 86 polymers was analyzed, and two quantitative structure–property relationship (QSPR) models were developed to predict dielectric permittivity. From an initial set of 1273 descriptors, the most relevant ones were selected using a genetic algorithm, and machine learning models were built using the Gradient Boosting Regressor (GBR). In contrast to Multiple Linear Regression (MLR)- and Partial Least Squares (PLS)-based models, the gradient boosting models excel in handling nonlinear relationships and multicollinearity, iteratively optimizing decision trees to improve accuracy without overfitting. The developed GBR models showed high R2 coefficients of 0.938 and 0.822, for the training and test sets, respectively. An Accumulated Local Effect (ALE) technique was applied to assess the relationship between the selected descriptors—eight for the GB_A model and six for the GB_B model, and their impact on target property. ALE analysis revealed that descriptors such as TDB09m had a strong positive effect on permittivity, while MLOGP2 showed a negative effect. These results highlight the effectiveness of the GBR approach in predicting the dielectric properties of polymers, offering improved accuracy and interpretability.

Ascencio-Medina, Estefania↗
DOE OSTI2025

Measurement of the inclusive isolated-photon production cross section in pp and Pb–Pb collisions at $\mathbf {\sqrt{\textit{s}_{NN }} = 5.02}$ TeV

The ALICE Collaboration at the CERN LHC has measured the inclusive production cross section of isolated photons at midrapidity as a function of the photon transverse momentum (p$^{γ}_{T}$), in Pb–Pb collisions in different centrality intervals, and in pp collisions, at centre-of-momentum energy per nucleon pair of $\sqrt{s_{NN}}$ = 5.02 TeV. The photon transverse momentum range is between 10–14 and 40–140 GeV/c, depending on the collision system and on the Pb–Pb centrality class. The result extends to lower p$^{γ}_{T}$ than previously published results by the ATLAS and CMS experiments at the same collision energy. The covered pseudorapidity range is |η γ | < 0.67. The isolation selection is based on a charged particle isolation momentum threshold p$^{iso, ch}_{T}$ GeV/c within a cone of radii R = 0.2 and 0.4. The nuclear modification factor is calculated and found to be consistent with unity in all centrality classes, and also consistent with the HG-PYTHIA model, which describes the event selection and geometry biases that affect the centrality determination in peripheral Pb–Pb collisions. The measurement is compared to next-to-leading order perturbative QCD calculations and to the measurements of isolated photons and Z 0 bosons from the CMS experiment, which are all found to be in agreement.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2024

Design and Optimization of Processes for Recovering Rare Earth Elements from End-of-Life Hard Disk Drives

In this poster, we first provide motivation for why rare earth elements as rare earth permanent magnets (REPM) are increasing in demand. We then highlight some of the recent work that has been done by several national labs (National Renewable Energy Laboratory (NREL), Environmental Protection Agency (EPA), Critical Minerals Institute (CMI)) on recycling rare earth elements from end-of-life hard disk drives (EOL). Then, we mention our long-term plan to design a feedstock agnostic process to recover rare earth elements as rare earth oxides from many different EOL products at once. Next, we discuss how we quantified the rare earth elements available for recycling from EOL hard disk drives from consumer desktops and laptops. We then discuss how we used superstructure optimization to design the optimal pathway. The proposed superstructure was modeled as a MILP optimization problem, selecting the net present value as the objective function. Costing data from the literature was used to inform this model whenever possible. However, due to the novelty of this research area, data were often unavailable, thus requiring the generation of flowsheets implemented in Aspen Plus.

Laliwala, Chris↗
DOE OSTI2026

Multi-trait multi-environment genomic prediction strategies for Miscanthus sacchariflorus

Genomic selection holds the potential to serve as a strategic tool to enhance the genetic gain of complex traits in Miscanthus breeding programs. The development of improved cultivars requires their assessment for various traits across diverse environments to ensure suitable overall performance. Hence, the multi-trait multi-environment (MTME) genomic prediction (GP) models offer an opportunity to improve selection accuracy. This study aims to evaluate the potential of five GP models: (1) three MTME models including genotype-by-trait-by-environment interaction (G×E×T) and (2) two single-trait multi-environment (STME) models (with and without G×E interaction). A Miscanthus sacchariflorus population comprising 336 genotypes evaluated in three environments and scored for four traits (biomass yield YDY, total culm number TCM, average internode length AIL, and culm node number CNN) was analyzed. The predictive ability of the models was evaluated considering three cross-validation schemes resembling realistic scenarios (CV1: predicting new genotypes, CVP: predicting missing traits in a given environment, and CV2: predicting partially observed genotypes). On average, in all cross-validation schemes compared to the STME the predictive ability of the MTME models was 10% to 70% higher for TCM and AIL. On the other hand, for YDY and CNN, both STME models performed similarly or slightly better (between 5 to 64%) than the MTME models in most environments. While the MTME models were not successful for all traits when compared to their STME counterparts, MTME models improved the prediction of the performance of genotypes that were untested across environments or lacked trait information in a specific environment. Overall, our study suggests that MTME GP models can be implemented in Miscanthus breeding programs to improve the predictive ability of the complex traits, shorten breeding cycles, and accelerate selection decisions.

genomic prediction (GP)↗
DOE OSTI2024

Non-Electricity Based Renewable Fuels: Theory and Computation for Solar Thermochemical Hydrogen

Dominated by photovoltaics and wind, current renewable energy sources generate mostly electricity, but 80% of the global final energy consumption occurs in form of fuels. Therefore, direct solar fuel generation would be a major breakthrough for the energy transition. Solar thermochemical hydrogen (STCH) is one of the very few potential routes towards scalable renewable fuels, but currently suffers from lack of an oxide working material that could optimally perform energy conversion within the thermodynamic boundary conditions. Theory and computation can contribute in two distinct ways, through materials search and discovery, but also by providing detailed mechanistic models for specific systems so to advance our understanding of possible design strategies. To enable high-throughput materials screening, we developed a defect graph neural network (dGNN) machine learning approach,[1] which accelerates the prediction of defect formation energies by replacing the tedious density functional theory (DFT) supercell calculations for all possible defect sites. This approach enables high-throughput database screening of oxides, which was integrated with thermodynamic modeling to extract the reduction entropies as additional selection criterion for STCH. Once potential candidate materials are identified, detailed models can guide materials design by predicting performance characteristics. One challenge is to quantitatively predict thermochemical equilibria at high concentrations when the redox active defects start to interact with each other, thereby impeding the formation of additional defects. Introducing a model for the free energy of defect interaction, parametrized on the basis of DFT data, we simulated the complete STCH redox cycle for (Sr,Ce)MnO3 alloys, achieving near-quantitative agreement with experimental data.[2] The analysis of these simulations reveals how defect interactions diminish the reduction entropy and H2 yield, suggesting to include these interactions in design considerations. Finally, we revisit the popular van't Hoff method for analyzing reduction enthalpies and entropies. This method is not ideal, as it involves a temperature-dependent convolution of gas-phase and solid-state entropies, causing uncertainties in the same order of magnitude as the physical quantities of interest. To avoid this problem, we suggest a simple alternative approach which can be applied to experimental and simulated data alike.

first-principles calculations↗
DOE OSTI2026

Maximizing machine learning interatomic potential transferability for the discovery of the novel stellated octadecagon Bi18-Pt24 cage structure

Achieving true transferability remains the central challenge for Machine Learning Interatomic Potentials (ML-IAPs) in modeling complex bimetallic nanoclusters across their vast potential energy surfaces. We systematically investigate data selection strategies to optimize the Chebyshev Interaction Model for Efficient Simulation (ChIMES) potential for the Bi-Pt nanoclusters by comparing three innovative sampling methods: Principal Component Analysis (PCA)/k-means (structural diversity), t-distributedStochasticNeighborEmbedding (t-SNE)/k-means (force-space diversity), and hierarchical clustering. Quantitatively, the PCA/k-means strategy proved most effective for global accuracy, yielding the lowest force errors and achieving energy root mean square errors (RMSE) values competitive with Density Functional Theory (DFT), demonstrating excellent accuracy (19.16meV/atom). Structural validation on 34 unique DFT-optimized isomers further confirmed the potential’s high fidelity, with the best model PCA/k-means reproducing structures with an average root mean square deviation (RMSD) of 0.10 Å. However, the t-SNE methods, by maximizing diversity in the force space, demonstrated superior extrapolative power, leading to the more precise prediction of a novel stellated octadecagon Bi18⁢Pt24 cage structure, demonstrating the potential for exploring previously unseen morphologies. Our results establish a clear methodology for strategic data sampling that successfully maximizes ML-IAP transferability, providing an accurate and computationally efficient tool that accelerates the theoretical discovery of complex bimetallic architectures.

Vangheluwe, Raphaël [Université Paris-Saclay, CNRS↗
DOE OSTI2024

Using ethanol and isopropanol as biomass model compounds for understanding bond scission mechanisms over Cu/Mo 2 N catalysts

The conversion of biomass compounds into fuels and chemicals is an important step towards a more sustainable future. This work combines results from model surfaces and powder catalysts to demonstrate Cu-modified mo- lybdenum nitride (Cu/Mo 2 N) as a selective catalyst for dehydrogenation of the biomass model compounds, ethanol and isopropanol. Results from model surfaces showed that while Mo2N led to unselective decomposition via both dehydrogenation and dehydration, the addition of Cu increased the dehydrogenation activity and selectivity. DFT calculations showed how Cu influenced the structures of active sites, adsorbate interactions, and thus the product selectivity. Batch reactor studies on corresponding powder catalysts confirmed the trend that Cu modification increased dehydrogenation activity, and in situ X-ray absorption spectroscopy elucidated the Cu oxidation state under reaction conditions. Further, this work demonstrates a strategy for promoting dehydrogenation over Mo 2 N-based catalysts, as well as the feasibility of using model surfaces to guide the design of industrially relevant catalysts.

09 BIOMASS FUELS↗
DOE OSTI2026

AutoBEM: A scalable framework for nationwide building energy simulation and retrofit evaluation in the United States

This paper presents AutoBEM, an integrated, automated framework for nationwide building energy modeling and retrofit evaluation in the United States. Unlike prior UBEM platforms that either rely primarily on representative stock sampling or operate at city scale, AutoBEM automates the generation of building-resolved, physics-based EnergyPlus/OpenStudio simulation models at national scale using GIS-derived geometry, prototype-based assumptions, and standardized scalable workflows. Leveraging the Model America dataset and high-performance computing, AutoBEM generates and simulates energy models for 122.9 million buildings, representing 97.8% of the U.S. building stock. These models are being made publicly and freely available as the Model America v1.0 (MAv1) dataset. AutoBEM supports detailed, building-level assessments of energy consumption, CO2 emissions, and post-processed anthropogenic heat emissions (AHE), and evaluates 151 energy conservation measures (ECMs) using localized utility pricing and building characteristics. In addition, AutoBEM incorporates both typical and future climate conditions through integration with Typical Meteorological Year (TMY) and Future TMY (fTMY) weather data derived from IPCC scenarios. In a case study of Phoenix, Arizona, AutoBEM identified several high-efficiency HVAC upgrades and selected envelope measures with short modeled payback periods (1.5 years) for certain building types and standards. Simulations under future climate scenarios (SSP5–RCP8.5) project an 11.3% increase in electricity use and a 32% reduction in natural gas demand by 2100, underscoring the need for climate-adaptive retrofit planning. By enabling reproducible, bottom-up, and location-specific analysis at scale, AutoBEM provides a step toward a national digital twin of the built environment and supports data-driven screening and planning for decarbonization, resilience, and energy equity.

Li, Hang [ORNL] (ORCID:0000000306001920)↗
DOE OSTI2024

Protein-Enabled Size-Selective Defect-Sealing of Atomically Thin 2D Membranes for Dialysis and Nanoscale Separations

Atomically thin 2D materials present the potential for advancing membrane separations via a combination of high selectivity (from molecular sieving) and high permeance (due to atomic thinness). However, the creation of a high density of precise nanopores (narrow-size-distribution) over large areas in 2D materials remains challenging, and nonselective leakage from nanopore heterogeneity adversely impacts performance. Here, we demonstrate protein-enabled size-selective defect sealing (PDS) for atomically thin graphene membranes over centimeter scale areas by leveraging the size and reactivity of permeating proteins to preferentially seal larger nanopores (≥4 nm) while preserving a significant amount of smaller nanopores (via steric hindrance). Our defect-sealed nanoporous atomically thin membranes (NATMs) show stability up to ~35 days during size-selective diffusive separations with a model dialysis biomolecule fluorescein isothiocyanate (FITC)-Ficoll 70 in phosphate buffer saline (PBS) solution as well as outperform state-of-the-art commercially available dialysis membranes (molecular-weight-cutoff ~3.5–5 kDa and ~8–10 kDa) with significantly higher permeance for smaller solutes KCl (~0.66 nm) ~5.1–6 × 10 –5 ms –1 and vitamin B12 (B12, ~1.5 nm) ~2.8–4 × 10 –6 ms –1 compared to small protein lysozyme (Lz, ~4 nm) ~4–6.4 × 10 –8 m s –1 , thereby allowing unprecedented selectivity for B12/Lz ~70 and KCl/Lz ~1280. Our work introduces proteins as nanoscale tools for size-selective defect sealing in atomically thin membranes to overcome persistent issues and advance separations for dialysis, protein desalting, small molecule separations/purification, and other bioprocesses.

77 NANOSCIENCE AND NANOTECHNOLOGY↗
DOE OSTI2025

Transient kinetic insights into selective propene oxidation over industrial bismuth molybdate catalysts

Selective oxidation of propene to acrolein over industrial multicomponent bismuth molybdate (BMO) catalysts significantly depends on reaction conditions that include operating parameters and catalyst state. Here, this work investigates selective propene oxidation in the intrinsic kinetic paradigm of Temporal Analysis of Products (TAP) reactor by systematically varying catalyst redox state, temperature, and oxygen-to-propene feed ratio. A 93 % propene conversion with an acrolein yield of 80 % is achieved at elevated temperatures (450 °C) on oxidized catalysts under oxygen-rich conditions (O 2 :C 3 H 6 = 10). However, these conditions diminish acrolein-to-CO 2 selectivity due to enhanced total oxidation to CO 2 . In contrast, a reduced catalyst state, moderate temperature (350 °C), and lower oxygen feed (O 2 :C 3 H 6 = 1) nearly doubles the acrolein to-CO 2 selectivity, albeit at a lower acrolein yield (33 %). Transient kinetic studies together with a kinetic model that simplify the major product formation pathways in lumped non-elementary forms reveal that the availability of surface oxygen species plays a pivotal role in governing reaction pathways. Additionally, density functional theory (DFT) calculations on pure BMO catalysts inform the role of surface redox states on propene and oxygen activation barriers. Lattice oxygen at acrolein-selective sites drives both acrolein and CO 2 formation, while adsorbed oxygen at activation sites favors unselective CO 2 generation. This work establishes a critical relationship between transient product selectivity and surface oxygen availability, which is strongly influenced by catalyst redox state, feed ratio, and reaction temperature. These insights underscore the importance of dynamic reactor operation strategies and offer a foundation for designing next-generation propene oxidation processes with tunable acrolein selectivity.

42 - ENGINEERING↗
DOE OSTI2025

Connecting relativistic density functional theory to microscopic calculations

The development of systematic effective field theories (EFTs) for nuclear forces and advances in solving the nuclear many-body problem have greatly improved our understanding of dense nuclear matter and the structure of finite nuclei. For global nuclear calculations, density functional theories (DFTs) have been developed to reduce the complexity and computational cost required in describing nuclear systems. However, DFT often makes approximations and assumptions about terms included in the functional, which may introduce systematic uncertainties compared to microscopic calculations using EFTs. In this work, we investigate possible avenues of improving nuclear DFT using nonlinear relativistic mean-field (RMF) theory. We explore the impact of RMF model extensions by fitting the nonlinear RMF model to predictions of nuclear matter and selected closed-shell nuclei using four successful chiral EFT Hamiltonians. We find that these model extensions are impactful and important in capturing the physics present within chiral Hamiltonians, particularly for charge radii and neutron skins of closed-shell nuclei. However, there are additional effects that are not captured within the RMF model, particularly within the isoscalar sector of RMF theory. Additional model extensions and the reliability of the nonlinear RMF model are discussed.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2025

Similarity Metric for Data Optimization and Efficient Training of Reactive Machine Learning Force Fields for Hydrocarbon Radiolysis

Radiolysis is a common approach to sterilize polymers, chemically modify them for upcycling, and accelerate their decomposition for recycling purposes. Reactive molecular dynamics (MD) simulations provide a powerful tool to generate atomic-level trajectories of the reactive processes and quantify radiolytic chemical degradation pathways. For this, machine learning (ML) surrogate models for reactive force fields with quantum mechanical accuracy are now widely used, which require ML training data sets that can provide information on atomic environments for target chemical systems. However, radiolysis chemistry can be highly complex and diverse, which poses significant challenges for generating training data to parametrize ML models. In this regard, we developed a method for optimizing the training data set using a cosine similarity metric to help guide training set selection for radiolysis of polyethylene, a model hydrocarbon polymer, as well as to enhance the transferability of our reactive ML force field (MLFF) to a variety of molecular and polymeric systems. Our approach performs atom-by-atom comparisons between local atomic environments to pinpoint important data points associated with rare and localized events, such as radiolysis damage within structures. We apply this approach to train the Chebyshev Interaction Model for Efficient Simulation (ChIMES) MLFF model, which expresses the atomic interaction potentials in terms of linear combinations of many-body Chebyshev polynomials. We first show that our method can reduce our training set size by ∼70% while improving overall accuracy compared to more standard MD model fitting approaches. We then validate our optimum model against diverse hydrocarbon simulation data, including simple alkanes and systems with unsaturated carbon bonds, over a wide range of thermodynamic conditions. Finally, we use our ChIMES model to perform MD simulations of radiolytic damage with large-scale systems that help avoid system size effects. Overall, our approach yields an MD force field that retains most of the accuracy of the underlying quantum method while yielding many orders of improvement in computational efficiency. In conclusion, our efforts will have impact on future hydrocarbon polymer radiolysis studies, where the chemical details of the polymer–radiation interactions can have a strong effect on the resulting products observed in experiments.

Hydrocarbons↗
DOE OSTI2026

Thermal Management for a Novel Non-Heavy Rare-Earth Interior Permanent Magnet Machine

The work presents a thermal management solution for a novel non-heavy rare-earth permanent magnet machine being developed at Oak Ridge National Laboratory. The motor has been designed to minimize losses while maximizing performance for a range of speeds and power ratings. The novel motor design reduces rare-earth magnet usage, thereby avoiding supply chain issues. The motor component heat losses are established for operating windows and desired performance. These heat losses, along with windage losses, are being used to develop cooling solutions for different components of this machine. A novel thermal management solution for stators and rotors has been developed, and progress is presented in this paper. The stator cooling is achieved with the help of water-ethylene glycol flowing over the finned aluminum stator jacket, and rotor cooling with automatic transmission fluid passing through novel channels designed in the rotor laminations. The attempt is to establish effective cooling of the stator winding, laminations, and rotor magnets. A 3D conjugate heat transfer model has been developed for overall thermal analysis to establish a down- selected thermal management solution for the machine. The model, in addition to estimated component heat losses, includes windage losses and its impact on rotor and stator cooling. Overall, the work presents a workable thermal solution for the interior permanent magnet machine with potential for further improvements. Future work will involve establishing end winding and refinement of other end parts of the machine with the aim of establishing a robust thermal management solution. The work will also focus on different shapes (e.g., round, non-round, presence of wedges) of rotor-stator gaps and investigate windage losses and their impact on thermal management for higher rotational speeds for the machine.

30 DIRECT ENERGY CONVERSION↗
DOE OSTI2025

On the minimum number of radiation field parameters to specify gas cooling and heating functions

Fast and accurate approximations of gas cooling and heating functions are needed for hydrodynamic galaxy simulations. We use machine learning to analyze atomic gas cooling and heating functions in the presence of a generalized incident local radiation field computed by Cloudy. We characterize the radiation field through binned radiation field intensities instead of the photoionization rates used in our previous work. We find a set of 6 energy bins whose intensities exhibit relatively low correlation. We use these bins as features to train machine learning models to predict Cloudy cooling and heating functions at fixed metallicity. We compare the relative SHapley Additive exPlanation (SHAP) value importance of the features. From the SHAP analysis, we identify a feature subset of 3 energy bins (0.5-1, 1-4, and 13-16Ry) with the largest importance and train additional models on this subset. We compare the mean squared errors and distribution of errors on both the entire training data table and a randomly selected 20% test set withheld from model training. The machine learning models trained with 3 and 6 bins, as well as 3 and 4 photoionization rates, have comparable accuracy everywhere, with errors ≳10 times smaller than for the interpolation table of Gnedin and Hollon (2012). We conclude that 3 energy bins (or 3 analogous photoionization rates: molecular hydrogen photodissociation, neutral hydrogen HI, and fully ionized carbon CVI) are sufficient to characterize the dependence of the gas cooling and heating functions on our assumed incident radiation field model.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2025

On the minimum number of radiation field parameters to specify gas cooling and heating functions

Fast and accurate approximations of gas cooling and heating functions are needed for hydrodynamic galaxy simulations. We use machine learning to analyze atomic gas cooling and heating functions computed by Cloudy in the presence of a generalized incident local radiation field. We characterize the radiation field through binned radiation field intensities instead of the photoionization rates used in our previous work. We find a set of 6 energy bins whose intensities exhibit relatively low correlation. We use these bins as features to train machine learning models to predict Cloudy cooling and heating functions at fixed metallicity. We compare the relative SHapley Additive exPlanation (SHAP) value importance of the features. From the SHAP analysis, we identify a feature subset of 3 energy bins ($0.5-1, 1-4$, and $13-16 \, \mathrm{Ry}$) with the largest importance and train additional models on this subset. We compare the mean squared errors and distribution of errors on both the entire training data table and a randomly selected 20% test set withheld from model training. The machine learning models trained with 3 and 6 bins, as well as 3 and 4 photoionization rates, have comparable accuracy everywhere, with errors $\gtrsim 10$ times smaller than for the interpolation table of Gnedin and Hollon (2012). We conclude that 3 energy bins (or 3 analogous photoionization rates: molecular hydrogen photodissociation, neutral hydrogen HI, and fully ionized carbon CVI) are sufficient to characterize the dependence of the gas cooling and heating functions on our assumed incident radiation field model.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2026

Selective Amnesia using Contrastive Subnet Erasure for Class Level Unlearning in Vision Models

We study concept-level forgetting in pretrained vision models: removing an entire semantic category so the system no longer recognizes that object in unseen images and contexts, rather than merely forgetting specific training examples. Prior work either applies blunt global projections or fine-tunes parameters, which can introduce collateral damage to unrelated features, add compute, and become unstable as forgetting strength increases. We introduce Contrastive Subnet Erasure (CSE), a training-free, encoder-centric edit that targets a compact set of channels most responsible for the class and attenuates them in a calibrated manner. The modification is algebraically folded into the subsequent layer, yielding no inference-time overhead and leaving task heads unchanged. To evaluate whether forgetting generalizes beyond the data used to specify the class, we introduce a cross dataset protocol in which the class is defined on a source dataset and performance is measured on a disjoint target dataset drawn from a different distribution with no shared images. This setup tests whether the model still fails to recognize the object when it looks different or appears in new scenes, and it helps avoid overfitting to patterns in the source dataset. Across CIFAR 10, CIFAR 100, and ImageNet under this protocol, CSE achieves stronger forgetting of the target class while better preserving non target utility than existing baselines in both single class and multi class settings. Overall, CSE provides a simple, stable, and deployment-ready mechanism for class-level unlearning in vision.

Kotevska, Olivera [ORNL] (ORCID:0000000316772243)↗