Search NASA⌕ Search

SEARCH · Search NASA

Results for “small dataset”

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 19 records
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

Probabilistic learning on manifolds constrained by nonlinear partial differential equations for small datasets

A novel extension of the Probabilistic Learning on Manifolds (PLoM) is presented. It makes it possible to synthesize solutions to a wide range of nonlinear stochastic boundary value problems described by partial differential equations (PDEs) for which a stochastic computational model (SCM) is available and which depend on a vector-valued random control parameter. The cost of a single numerical evaluation of this SCM is assumed to be such that only a limited number of points can be computed for constructing the training dataset (small data). Each point of the training dataset is made up of realizations from a vector-valued stochastic process (the stochastic solution) and the associated random control parameter on which it depends. The presented PLoM constrained by PDE allows for generating a large number of learned realizations of the stochastic process and its corresponding random control parameter. These learned realizations are generated so as to minimize the vector-valued random residual of the PDE in the mean-square sense. Appropriate novel methods are developed to solve this challenging problem. Three applications are presented. The first one is a simple uncertain nonlinear dynamical system with a nonstationary stochastic excitation. The second one concerns the 2D nonlinear unsteady Navier–Stokes equations for incompressible flows in which the Reynolds number is the random control parameter. Here, the last one deals with the nonlinear dynamics of a 3D elastic structure with uncertainties. The results obtained make it possible to validate the PLoM constrained by stochastic PDE but also provide further validation of the PLoM without constraint.

Machine learning↗
DOE OSTI2025

Applying Deep Learning for Wildfire Identification: Economical and Accessible Solutions Leveraging Small Datasets

Wildfires significantly impact human health, air quality, visibility, weather, and climate change and cause substantial economic losses. While state and county-operated air quality monitors provide critical insights during wildfires, they are not available in all regions. This highlights the need for affordable, accessible tools that allow the general public to assess air quality impacts. In this study, we apply machine learning with deep neural networks to diagnose air quality rapidly from sky images taken at the Pacific Northwest National Laboratory in Richland, WA, USA. Using a convolutional neural network (CNN) framework, we trained a deep learning model to classify air quality indices based on sky images. By leveraging transfer learning, our approach fine-tunes a pre-trained model on a small dataset of sky images, significantly reducing training time while maintaining high accuracy. Our results demonstrate the potential of deep learning to provide rapid air quality diagnostics during wildfire episodes, offering early warnings to the public and enabling timely mitigation strategies, particularly for vulnerable populations. Additionally, we show that lower respiratory infections pose the highest health risk during acute smoke exposures. Reactive oxygen species (ROS) from wildfire particles further exacerbate health risks by triggering inflammation and other adverse effects.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

Probabilistic-learning-based stochastic surrogate model from small incomplete datasets for nonlinear dynamical systems

We consider a high-dimensional nonlinear computational model of a dynamical system, parameterized by a vector-valued control parameter, in the presence of uncertainties represented by an uncontrolled parameter modeled by a vector-valued random variable, and possibly with stochastic excitation. The objective is to construct a statistical surrogate model where the input is any deterministic value of the control parameter, and the output is a vector-valued observation of the computational model, which is a random vector whose probability measure is updated using a target dataset. To construct this statistical surrogate model, the stochastic response of the computational model must be built, which is a vector-valued time-discretized stochastic process in high dimension, depending on the control parameter. It is assumed that the computational cost of a single evaluation of the deterministic model is high. For the probabilistic updating, we consider a subset of the components of the observation of the computational model, defined as the “identification observation” of the computational model, for which a small target dataset is available. Therefore, the target dataset is associated with partial observability, corresponding to an incomplete data case. Given a prior probability model of the random control and uncontrolled parameters, a training dataset is constructed, consisting of realizations of the random triplet composed of the stochastic response, the random identification observation, and the random control parameter. Since the computational cost of a single evaluation of the deterministic model is assumed to be large, the training dataset is also of small size. The main challenges in this problem are the high dimensionality, partial observability leading to incomplete data in the target dataset for the identification observation of the computational model (which is not sufficient to identify the computational stochastic responses), and the availability of a small training dataset. To address these challenges, we propose a methodology based on statistical methods for constructing necessary reduced representations, direct probabilistic learning under constraints using probabilistic learning on manifolds (PLoM) constrained by the target dataset, and the use of a weak formulation of the Fourier transform of probability measures. Statistical conditioning is also employed to explore the learned dataset. The constructed predictive statistical surrogate model can be implemented in the context of online computation. Here, we apply this approach to a problem of nonlinear stochastic dynamics in high dimensions within the framework of deformable solids mechanics.

Engineering↗
DOE OSTI2019

The METLIN small molecule dataset for machine learning-based retention time prediction

Machine learning has been extensively applied in small molecule analysis to predict a wide range of molecular properties and processes including mass spectrometry fragmentation or chromatographic retention time. However, current approaches for retention time prediction lack sufficient accuracy due to limited available experimental data. Here we introduce the METLIN small molecule retention time (SMRT) dataset, an experimentally acquired reverse-phase chromatography retention time dataset covering up to 80,038 small molecules. To demonstrate the utility of this dataset, we deployed a deep learning model for retention time prediction applied to small molecule annotation. Results showed that in 70% of the cases, the correct molecular identity was ranked among the top 3 candidates based on their predicted retention time. We anticipate that this dataset will enable the community to apply machine learning or first principles strategies to generate better models for retention time prediction.

59 BASIC BIOLOGICAL SCIENCES↗
NASA NTRSDate not supplied

Watching Without Seeing a Tool to Surveil Astronaut Health Outcomes While Maintaining Astronaut Medical Privacy

BACKGROUND The Privacy Act of 1974 regulates the use a nd disclosure of personally identifiable information by US Federal agencies. The Act applies to biographical, financial, a nd other identity-linked information, a s well a s personal health information (PHI). As such, the use of astronaut PHI is limited to authorized personnel for preapproved uses, with data reporting often limited to aggregated information about groups. These limitations on the use a nd reporting of astronaut PHI complicates surveillance efforts, wherein epidemiologists a t the National Aeronautics and Space Administration (NASA)monitor the incidence of targeted health conditions in the astronaut population, or to discover emerging trends of aging and disease. Stratification on one or more covariates –particularly time-period, sex, a nd mission participation –can lead to extremely small datasets such that the reporting of results is potentially attributable to individuals. An additional challenge is the small size of the astronaut population, both in terms of numbers of individuals a s well a s in terms of density of exposure time. Such small datasets yield volatile rate estimates that are difficult to interpret. To a id the epidemiological surveillance efforts, a surveillance tool is required that can (a) satisfy the need for rapid computation of condition-specific incidence and mortality rates; (b) improve the statistical estimates of these estimated rates; and (c) maintain astronaut privacy. Here we describe a nd demonstrate such a tool. METHODS We devised a system that models incidence a nd mortality rates rather than calculating them directly. This ha s the advantage of using all the available data to derive the estimates, lea ding to rates that a re not attributable to any one individual, a nd a re a s numerically stable a s they can be given the extremely limited data. The system models disease endpoints using a Poisson regression model with exposure density (measured in person-years) a s a n offset term. By doing so the model is estimating event counts per person-year, equivalent to modeling the rates directly. It uses a standard (pre-specified)set of covariates; the system does not engage in “model-building” as model parsimony is not the goa l. Instead, it is explicitly recognized that if a covariate is not statistically significant a nd not a confounder then it will likely have very little effect on the estimate of the incidence a nd mortality rates. Users are able to specify the disease endpoint of interest and the covariates over which they would like to stratify. The system then uses the resulting model to compute the estimated rates for the user-chosen configuration of variables as visualizes those either over an age range within a specified time-period, or over time for astronauts with a specified age range. RESULTS The first iteration of the tool computes incidence a nd mortality rates for cardiovascular conditions and cancers. Code ha s been developed to retrieve the appropriate data from the IMPALA analysis platform, compute the models for incidence a nd mortality, a nd then use those models to generate the corresponding rate curves. A companion graphical user interface allows the user to specify the curves and visualize the results. CONCLUSIONS It is important to note that the rapid surveillance tool described here is neither meant to be a definitive assessment of the incidence or mortality of any particular disease or condition in the astronaut population, nor is it meant to be used for research purposes. Rather, it is meant as an early indicator that in-depth investigation may be warranted. By automating a repetitive process and leveraging carefully curated astronaut health outcomes, the tool makes possible a rapid “first look” into known areas of concern, and, if used judiciously, may surface new areas of concern for long-term astronaut health. This work is supported in part by the Translational Research Institute for Space Health (TRISH) through NASA Cooperative Agreement NNX16AO69A.

R J Reynolds↗
DOE OSTI2022

Synthetic Data Generation Using Machine Learning

Robust machine learning techniques for image analysis require a substantial amount of data to yield confident results. In the nuclear domain, data scarcity is a substantial challenge because there are so few facilities worldwide. This research focuses on being able alleviate the data scarcity problem by generating synthetic data to bridge the gap between large and small datasets. This work achieves that goal using a Generative Adversarial Network (GAN) architectural approach, by training a model on real-world data and expands that small dataset through synthetic data amendments. Model performance is impacted by the size of the real-world dataset and the number of training epochs utilized. This means that 1) It is important to develop your GAN to be optimized with the specific data type, and 2) approaches taken when training the GAN should be specialized to encompass important aspects of the dataset that it is generating. By taking a step to improve dataset sizes in this way, the gap between models trained by parties with significant amount of data and those without access to large data, closes, allowing for robust analyses of satellite imagery for nuclear domain applications.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
NASA NTRS2024

Assessments of Physiology and Cognition in Hybrid-Reality Environments (APACHE) – A VR “Sandbox” for Planetary Spacewalk Research

NASA is planning to return to the Moon in the mid-2020s as a steppingstone to Mars missions in the 2030s. Spacewalks, or extravehicular activities (EVAs), performed on the Moon and Mars will differ in a variety of ways from those that have been performed in decades past. NASA has identified multiple risks to human health and performance associated with a crewed mission to Mars, especially those associated with exploration EVAs which are expected to be a primary mission activity. Crew may be expected to conduct up to 24 hours of EVA per person per week, where the likelihood of injury and/or mental mistakes are increased compared to ground-based training or current microgravity EVAs and the consequences of which can be catastrophic. Current test environments for exploration EVA research and technology development are large, costly facilities that are limited in their availability or capabilities. Spacesuit testing in a reduced gravity environment such as NASA’s Neutral Buoyancy Laboratory, while a good representation of the crew’s physical workload during exploration EVAs, typically has small datasets and is difficult to integrate physiological sensors or other types of crew performance measures. Meanwhile, scientific field-based testing such as NASA’s Desert Research and Technology Studies offers an operationally relevant environment for exploration EVAs, particularly for cognitive workload, but is also limited by small datasets, lack of a pressurized spacesuit, and obtrusive measures. The limitations of current analogs for exploration EVAs identify a need for a new test environment that can approximate both the physical and cognitive demands associated with exploration EVAs to enable rapid, controlled, and repeatable evaluations of human health and performance risks of exploration missions. In response, the Human Physiology, Performance, Protection, and Operations Laboratory (H-3PO) at NASA Johnson Space Center has developed a hybrid reality exploration EVA analog named the Assessments of Physiology And Cognition in Hybrid-reality Environments (APACHE)[ to address these limitations using a combination of virtual, physical, and hybrid reality techniques. The APACHE facility resides at NASA Johnson Space Center and serves as a large “sandbox” for EVA research and simulation. At its center is a roughly 15x20ft space surrounded by a 14” tall sandbox partially filled with lunar regolith simulant to emulate the physical feeling of walking on a planetary surface and to allow for simulated geology operations. Nearby, a curved passive treadmill (Skillmill Connect, Technogym, Fairfield, NJ) and an omnidirectional treadmill (Infinadeck, Infinadeck, Rocklin, CA) are included to enable exploration of these large virtual environments while also imposing the physical demands, representative timelines, and cognitive burdens required to navigate and traverse these distances during exploration EVA. A 6DOF motion platform is used to simulate rover operations and supports various human performance evaluations and associated risks. Lastly, APACHE can support two extravehicular (EV) crewmembers working in tandem. A computer workstation is located nearby and also supports an intravehicular (IV) crewmember as part of a full mission simulation. The IV crewmember has direct video and audio communication with the EV crew in VR to provide operational and procedural support.

Human Performance↗
NASA NTRSDate not supplied

Assessments of Physiology and Cognition in Hybrid-Reality Environments (APACHE)

NASA is planning to return to the Moon in the mid-2020s as a stepping stone to Mars missions in the 2030s. Spacewalks, or extravehicular activities (EVAs), performed on the Moon and Mars will differ in a variety of ways from those that have been performed in decades past. NASA has identified multiple risks to human health and performance associated with a crewed mission to Mars, especially those associated with exploration EVAs which are expected to be a primary mission activity. Crew may be expected to conduct up to 24 hours of EVA per person per week, where the likelihood of injury and/or mental mistakes are increased compared to ground-based training or current microgravity EVAs and the consequences of which can be catastrophic. Current test environments for exploration EVA research and technology development are large, costly facilities that are limited in their availability or capabilities. Spacesuit testing in a reduced gravity environment such as NASA’s Neutral Buoyancy Laboratory, while a good representation of the crew’s physical workload during exploration EVAs, typically has small datasets and is difficult to integrate physiological sensors or other types of crew performance measures. Meanwhile, scientific field-based testing such as NASA’s Desert Research and Technology Studies offers an operationally relevant environment for exploration EVAs, particularly for cognitive workload, but is also limited by small datasets, lack of a pressurized spacesuit, and obtrusive measures. The limitations of current analogs for exploration EVAs identify a need for a new test environment that can approximate both the physical and cognitive demands associated with exploration EVAs to enable rapid, controlled, and repeatable evaluations of human health and performance risks of exploration missions. In response, the Human Physiology, Performance, Protection, and Operations Laboratory (H-3PO) at NASA Johnson Space Center has developed a hybrid reality exploration EVA analog named the Assessments of Physiology And Cognition in Hybrid-reality Environments (APACHE) to address these limitations using a combination of virtual, physical, and hybrid reality techniques. The APACHE facility resides at NASA Johnson Space Center and serves as a large “sandbox” for EVA research and simulation. At its center is a roughly 15x20ft space surrounded by a 14” tall sandbox partially filled with lunar regolith simulant to emulate the physical feeling of walking on a planetary surface and to allow for simulated geology operations. Nearby, a curved passive treadmill (Skillmill Connect, Technogym, Fairfield, NJ) and an omnidirectional treadmill (Infinadeck, Infinadeck, Rocklin, CA) are included to enable exploration of these large virtual environments while also imposing the physical demands, representative timelines, and cognitive burdens required to navigate and traverse these distances during exploration EVA. A 6DOF motion platform is used to simulate rover operations and supports various human performance evaluations and associated risks. Lastly, APACHE can support two extravehicular (EV) crewmembers working in tandem. A computer workstation is located nearby and also supports an intravehicular (IV) crewmember as part of a full mission simulation. The IV crewmember has direct video and audio communication with the EV crew in VR to provide operational and procedural support. The software used in APACHE was created by the JSC Engineering Directorate, in partnership with Buendea, powered by a custom Unreal Engine 5 (UE5.3, Epic Games) project. APACHE currently utilizes the HTC Vive Pro Eye in a wireless configuration for VR simulations. There are two virtual environments that subjects can explore within APACHE, a Lunar and Martian surface. The virtual Lunar surface was created from LIDAR data of the Lunar South Pole to create roughly 16 sq km of explorable terrain. The virtual Martian surface contains roughly 400 sq km of explorable terrain derived from Mars Reconnaissance Orbiter LIDAR data of the Jezero Crater. The immersion and related cognitive burdens of conducting a planetary EVA is simulated through a series of EVA-relevant tasks performed in the VR environment, using these high-fidelity visual representations. Additionally, APACHE includes biosensor driven informatics, such as real-time heart rate monitoring and/or derived values from model simulations, for active monitoring by the EV crew and added cognitive demand. A “Wizard of Oz” control panel enables test operators to activate contingency events such as simulated spacesuit malfunctions, loss of communications, and/or limited visibility. Embedded performance measures such as accuracy, completeness, and execution time have been developed for various exploration tasks to objectively quantify crew performance during an EVA and compare impacts to performance when different environmental stressors, both physical and cognitive, are added to or removed from the simulation. Additionally, validated cognitive and operational performance measures such as the Digit Symbol Substitution Task have been recreated and embedded in VR for direct and relatively unobtrusive measurement of motor perception. The APACHE environment currently supports multiple research studies at NASA. Examples include the CHAPEA project, a series of simulated year-long missions on Mars by a 4-person crew; and the CO2 Contingency Walk Back Study, an investigation of elevated CO2 exposure on crew performance during a contingency EVA scenario. APACHE also provides a test environment to support the development of the Crew State and Risk Model, which is a collection of individualized, mathematical models of crew physical and cognitive state; and the Personalized EVA Informatics and Decision Support system, an operational tool for flight controllers, and eventually a self-reliant Martian crew, to make biomedically-informed decisions in real-time to optimize the EVA planning and execution with respect to crew health and performance. Some technical challenges associated with developing the APACHE environment, as well as current limitations, include VR limitless natural walking with a hybrid spacesuit simulator, optimizing performance for wireless PC VR streaming while maintaining a high degree of visual fidelity, and the integration of various physiological (metabolic masks) and psychometric (eye tracking) sensors with the VR headset.

Human Performance↗
DOE OSTI2020

Imprecise global sensitivity analysis using bayesian multimodel inference and importance sampling

Global Sensitivity Analysis (GSA) aims to understand the relative importance of uncertain input variables to model response. Conventional GSA involves calculating sensitivity (Sobol’) indices for a model with known model parameter distributions. However, model parameters are affected by aleatory and epistemic uncertainty, with the latter often caused by lack of data. In this paper, we propose a new framework to quantify uncertainty in probability model-form and model parameters resulting from small datasets and integrate these uncertainties into Sobol’ index estimates. First, the process establishes, through Bayesian multimodel inference, a set of candidate probability models and their associated probabilities. Imprecise Sobol’ indices are calculated from these probability models using an importance sampling reweighting approach. This results in probabilistic Sobol’ indices, whose distribution characterizes uncertainty in the sensitivity resulting from small dataset size. The imprecise Sobol’ indices thus provide a measure of confidence in the sensitivity estimate and, moreover, can be used to inform data collection efforts targeted to minimize the impact of uncertainties. Through an example studying the parameters of a Timoshenko beam, we show that these probabilistic Sobol’ indices converge to the true/deterministic Sobol’ indices as the dataset size increases and hence, distribution-form uncertainty reduces. The approach is then applied to assess the sensitivity of the out-of-plane properties of an E-glass fiber composite material to its constituent properties. This second example illustrates the approach for an important class of materials with wide-ranging applications when data may be lacking for some input parameters.

42 ENGINEERING↗
DOE OSTI2020

Overcoming small minirhizotron datasets using transfer learning

Minirhizotron technology is widely used to study root growth and development. Yet, standard approaches for tracing roots in minirhiztron imagery is extremely tedious and time consuming. Machine learning approaches can help to automate this task. However, lack of enough annotated training data is a major limitation for the application of machine learning methods. Transfer learning is a useful technique to help with training when available datasets are limited. In this paper, we investigated the effect of pre-trained features from the massives-cale, irrelevant ImageNet dataset and a relatively moderate-scale, but relevant peanut root dataset on switchgrass root imagery segmentation applications. We compiled two minirhizotron image datasets to accomplish this study: one with 17,550 peanut root images and another with 28 switchgrass root images. Both datasets were paired with manually labeled ground truth masks. Deep neural networks based on the U-net architecture were used with different pre-trained features as initialization for automated, precise pixel-wise root segmentation in minirhizotron imagery. We observed that features pre-trained on a closely related but relatively moderate size dataset like our peanut dataset were more effective than features pre-trained on the large but unrelated ImageNet dataset. Here, we achieved high quality segmentation on peanut root dataset with 99.04% accuracy at the pixel-level and overcame errors in human-labeled ground truth masks. By applying transfer learning technique on limited switchgrass dataset with features pre-trained on peanut dataset, we obtained 99% segmentation accuracy in switchgrass imagery using only 21 images for training (fine tuning). Furthermore, the peanut pre-trained features can help the model converge faster and have much more stable performance.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2022

Monotonic Gaussian Process for Physics-Constrained Machine Learning With Materials Science Applications

Physics-constrained machine learning is emerging as an important topic in the field of machine learning for physics. One of the most significant advantages of incorporating physics constraints into machine learning methods is that the resulting model requires significantly less data to train. By incorporating physical rules into the machine learning formulation itself, the predictions are expected to be physically plausible. Gaussian process (GP) is perhaps one of the most common methods in machine learning for small datasets. In this paper, we investigate the possibility of constraining a GP formulation with monotonicity on three different material datasets, where one experimental and two computational datasets are used. The monotonic GP is compared against the regular GP, where a significant reduction in the posterior variance is observed. The monotonic GP is strictly monotonic in the interpolation regime, but in the extrapolation regime, the monotonic effect starts fading away as one goes beyond the training dataset. Imposing monotonicity on the GP comes at a small accuracy cost, compared to the regular GP. The monotonic GP is perhaps most useful in applications where data are scarce and noisy, and monotonicity is supported by strong physical evidence.

36 MATERIALS SCIENCE↗
DOE OSTI2021

ChemPix: automated recognition of hand-drawn hydrocarbon structures using deep learning

Inputting molecules into chemistry software, such as quantum chemistry packages, currently requires domain expertise, expensive software and/or cumbersome procedures. Leveraging recent breakthroughs in machine learning, we develop ChemPix: an offline, hand-drawn hydrocarbon structure recognition tool designed to remove these barriers. A neural image captioning approach consisting of a convolutional neural network (CNN) encoder and a long short-term memory (LSTM) decoder learned a mapping from photographs of hand-drawn hydrocarbon structures to machine-readable SMILES representations. We generated a large auxiliary training dataset, based on RDKit molecular images, by combining image augmentation, image degradation and background addition. Additionally, a small dataset of ~600 hand-drawn hydrocarbon chemical structures was crowd-sourced using a phone web application. These datasets were used to train the image-to-SMILES neural network with the goal of maximizing the hand-drawn hydrocarbon recognition accuracy. By forming a committee of the trained neural networks where each network casts one vote for the predicted molecule, we achieved a nearly 10 percentage point improvement of the molecule recognition accuracy and were able to assign a confidence value for the prediction based on the number of agreeing votes. The ensemble model achieved an accuracy of 76% on hand-drawn hydrocarbons, increasing to 86% if the top 3 predictions were considered.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2020

A Bayesian framework for adsorption energy prediction on bimetallic alloy catalysts

Abstract For high-throughput screening of materials for heterogeneous catalysis, scaling relations provides an efficient scheme to estimate the chemisorption energies of hydrogenated species. However, conditioning on a single descriptor ignores the model uncertainty and leads to suboptimal prediction of the chemisorption energy. In this article, we extend the single descriptor linear scaling relation to a multi-descriptor linear regression models to leverage the correlation between adsorption energy of any two pair of adsorbates. With a large dataset, we use Bayesian Information Criteria (BIC) as the model evidence to select the best linear regression model. Furthermore, Gaussian Process Regression (GPR) based on the meaningful convolution of physical properties of the metal-adsorbate complex can be used to predict the baseline residual of the selected model. This integrated Bayesian model selection and Gaussian process regression, dubbed as residual learning, can achieve performance comparable to standard DFT error (0.1 eV) for most adsorbate system. For sparse and small datasets, we propose an ad hoc Bayesian Model Averaging (BMA) approach to make a robust prediction. With this Bayesian framework, we significantly reduce the model uncertainty and improve the prediction accuracy. The possibilities of the framework for high-throughput catalytic materials exploration in a realistic setting is illustrated using large and small sets of both dense and sparse simulated dataset generated from a public database of bimetallic alloys available in Catalysis-Hub.org.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2024

Assessment of fine-tuned large language models for real-world chemistry and material science applications

The current generation of large language models (LLMs) has limited chemical knowledge. Recently, it has been shown that these LLMs can learn and predict chemical properties through fine-tuning. Using natural language to train machine learning models opens doors to a wider chemical audience, as field-specific featurization techniques can be omitted. In this work, we explore the potential and limitations of this approach. We studied the performance of fine-tuning three open-source LLMs (GPT-J-6B, Llama-3.1-8B, and Mistral-7B) for a range of different chemical questions. We benchmark their performances against “traditional” machine learning models and find that, in most cases, the fine-tuning approach is superior for a simple classification problem. Depending on the size of the dataset and the type of questions, we also successfully address more sophisticated problems. The most important conclusions of this work are that, for all datasets considered, their conversion into an LLM fine-tuning training set is straightforward and that fine-tuning with even relatively small datasets leads to predictive models. These results suggest that the systematic use of LLMs to guide experiments and simulations will be a powerful technique in any research study, significantly reducing unnecessary experiments or computations.

Van Herck, Joren↗
DOE OSTI2026

Citation network datasets for benchmarking spiking graph neural networks on experimental neuromorphic hardware

Spiking neural networks (SNNs) running on neuromorphic computers offer an energy-efficient alternative for AI tasks. Recently, spiking graph neural networks (S-GNNs) have been shown to produce encouraging results on benchmark citation network datasets such as Cora, CiteSeer, and PubMed for node classification tasks. These S-GNNs were run on SNN simulators only because they contain up to tens of thousands of neurons and up to millions of synapses, translating poorly to neuromorphic hardware. Therefore, in this paper, we create a suite of benchmark datasets from the CiteSeer dataset that can be accommodated on current neuromorphic hardware platforms. Our contribution consists of a collection of three datasets. First, we have an induced subgraph of CiteSeer, which we call MiniSeer, containing 2110 papers, 3604 binary features, and 6 topics. Second, MicroSeer is a very small dataset consisting of 84 papers, 1227 features, and 6 topics. Lastly, BiteSeer is a collection of 15 binary classification datasets. We present creation of these datasets along with accuracies, running times, and spike counts when simulated. We believe that our results in this paper will be used by the neuromorphic community to benchmark, test, and develop neuromorphic hardware and simulators.

Zhu, Kevin [George Mason University, Virginia]↗
DOE OSTI2025

Antiviral discovery using sparse datasets by integrating experiments, molecular simulations, and machine learning

Computational methods have demonstrated success in identifying virucidal agents, effectively contributing to the discovery of novel virucidal molecules. In this study, we developed a machine learning (ML) model, trained on a small dataset, to predict inhibitors of human enterovirus 71 (EV71), a pathological agent that causes severe disease in children and immunocompromised adults. Despite the dataset’s limitation, comprising of only 36 compounds tested, our ML framework demonstrated significant predictive capability. Notably, experimental validation revealed that five out of the eight compounds predicted by our model from the Chinese cosmetic material list exhibited virucidal activity. The inhibitor effects displayed by the main active compounds were further confirmed by molecular dynamics simulation. This underscores the potential of our AI-driven approach to bypass data constraints in identifying active molecules against viral pathogens.

60 APPLIED LIFE SCIENCES↗