Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Decision tree”

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 379 records · Page 21
All sourcesNASA NTRSDOE OSTI
DOE OSTI2020

Attack-resilient algorithms and testbed federation for wide-area protection and control in smart grid

Today’s energy infrastructure is undergoing a massive transformation across all generation, transmission, and distribution systems to provide reliability, efficiency, and sustainability to the power system network. During the past 10 years, there has been a remarkable enhancement in developing close-loop wide-area protection and control (WAPAC) applications, such as automatic generation control (AGC), wide-area protection scheme (WAPS), synchrophasor-based wide-area voltage control system (WAVCS), etc., in the energy management system (EMS) to provide an adequate situational awareness to control center operators during physical disturbances and natural events. The rapid digitalization, unencrypted communication, and numerous data-sharing devices have increased digital access points that have exposed the grid network to countless cyberattacks pertinent to the WAPAC cybersecurity. Given the amount of complex grid network coupled with legacy infrastructures, it is untrustworthy to completely rely on information technology (IT)-based conventional security measures to defend against evolving cyberattacks. Hence, there is a compelling need to go beyond the traditional paradigm of ‘security by obscurity’ and ‘bolt-on’ security measures, and develop a suite of innovative security solutions at multiple layers to transform the current “fault-resilient” grid into “fault and attack-resilient” grid of the future. This dissertation specifically focuses on developing application-specific attack-resilient solutions for critical WAPAC applications in the smart grid. The first part of the study, Attack-Resilient Wide-Area Protection Scheme, presents a cyber-physical attack-resilient system (CPARS) for WAPS, where two anomaly detection methods are proposed to detect multi-level cyberattacks followed by the rules-based intrusion mitigation system (RIMS) to initiate effective mitigation strategies, tailored to these attacks, to restore the normal grid operation after disturbances. The first anomaly detection component, Machine learning-based Anomaly Detection for Centralized WAPS, discusses a novel architecture and methodology for developing a machine-learning-based anomaly detection system (ADS) for the centralized wide-area protection system (CWAPS) by applying the variational mode decomposition (VMD) technique and decision tree (DT) algorithms using cyber logs and synchrophasor measurements. The second component, Multi-Agent-based Anomaly Detection for Decentralized WAPS, presents a two-level hierarchical multi-agent-based decentralized WAPS against the system-aware stealthy cyber-attacks. Finally, for the RIMS, the state transition diagram-based mitigation measures are defined based on the grid operation states and experimental testing and evaluation are presented in a cyber-physical testbed environment. The second part of the study, Attack-Resilient Wide-Area Voltage Control System, addresses the existing vulnerabilities in the wide-area voltage control system (WAVCS), presents the methodology for performing an impact analysis during data integrity attacks, and proposes a data-driven attackresilient system (DARS) by incorporating machine learning-based ADS and rules-based anomaly mitigation system (RAMS) to detect these attacks and provide necessary mitigation actions to maintain the transient voltage stability after disturbances. Finally, the third part, Networked Federation Testbed (NEFTSec) for Smart Grid Cybersecurity, focuses on developing an interconnected federated testbed through a common network like the internet, describes its conceptual architectures, and also presents its design components for experimental testing, validation, and evaluation. Further, it presents the co-simulation interface algorithm (CIA) to facilitate the geographically-dispersed real-time distributed simulation (GD-RTDS) during the cyber-physical federation. In general, this dissertation presents innovative attack-resilient solutions for WAPAC applications that are experimentally tested and evaluated using the PowerCyber testbed and NEFTSec platform to accelerate the transition of state-of-the-art research works to the real-field deployments.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2023

Large Language Models (LLMs) for Energy Systems Research

The integration of Large Language Models (LLMs) in energy systems research promises transformative results, as demonstrated in this work, particularly in the realms of information retrieval and legal document analysis. We have developed a chat-based interface, specifically designed to query an extensive corpus of technical reports from the National Renewable Energy Laboratory (NREL). This interface capitalizes on the natural language processing capabilities of LLMs, providing future consumers of NREL research with a user-friendly platform to access and extract valuable information from technical documents, thus enhancing the dissemination of research to the public. In addition to information retrieval, we have employed LLMs to extract renewable energy siting ordinances from a variety of legal documents, a task traditionally driven by significant human labor. This automated extraction not only supports the ongoing development of the high-impact NREL siting ordinance database but also ensures the database's accuracy and comprehensiveness. Crucially, we have augmented the performance of LLMs through the integration of a decision tree framework, resulting in a substantial improvement in extraction accuracy. Comparative analysis with manual efforts has shown that this approach not only rivals but also significantly surpasses human accuracy, heralding increased reliability in legal document analysis for energy systems research. To democratize access to these advancements and foster collaborative research, we introduce the "Energy Language Model" (ELM), an open-source software package. ELM encapsulates the methodologies and tools developed in this work, providing researchers and practitioners with a robust toolkit to conduct similar analyses within their respective domains. Through these contributions, this work underscores the immense potential of LLMs in revolutionizing energy systems research, improving accuracy, efficiency, and accessibility in the field.

automation↗
DOE OSTI2023

A Search for Heavy Neutral Leptons Decaying to Electron-Positron Pairs and Neutral Pions in the MicroBooNE Detector

This thesis presents a search for heavy neutral leptons (HNLs) decaying inside the MicroBooNE detector. The MicroBooNE detector is a liquid-argon time projection chamber located at Fermilab that receives flux from two neutrino beams; the booster neutrino beam (BNB) and the neutrinos at the main injector (NuMI) beam. The search uses data taken when the NuMI beam was on, with a total exposure of $7.01\times10^{20}$ protons on target. HNLs would be produced in kaon decays, and the source of HNLs targeted in this search is kaons decaying at rest in the NuMI hadron absorber. The kinematics of the HNLs produced in this way allow their decays to be distinguished from the majority of beam-neutrino interactions and cosmic-ray activity that constitute the background. Two HNL decay channels are studied, $\nu$\ee and \vpi. Boosted decision trees (BDTs) are trained to separate the simulated HNL decays from the expected background events. The BDT score distributions are used to calculate upper limits on the mixing angle \umusq in the mass range $10\le m_{\rm HNL}\le 245$~MeV. This is the first direct search for HNL decays to the \vpi channel, and the limits provide the most stringent constraints on \umusq in the mass range $35<\mhnl<175$~MeV.

43 PARTICLE ACCELERATORS↗
DOE OSTI2023

Annular Metallic Nuclear Fuel Informatics at 50 nm Resolution

U-10wt.% Zr (U-10Zr) based metallic fuel is the leading candidate for next-generation sodium cooled fast reactor in United States. Advanced post-irradiation characterization (from sub-nanometer to micrometer) helps to understand fuel microstructure and property change during irradiation, benefiting fuel qualification for commercial application. With high velocity image data generating method, an automatic way to extract the microstructural information quantitively can better serve the needs from post irradiation characterization. A trained machine learning model, named Decision Tree, is employed to categorize pores caused by fission gas release and to aid phase identification. This work presents a showcase of this approach on different irradiated U-10Zr metallic fuels. This quantitative data offers insights into the fission product migration and potentially thermal conductivity degradation. This information from machine learning will be fed into fuel design code for better prediction of fuel performance.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
DOE OSTI2024

Emerging Jets Search, Triton Server Deployment, and Track Quality Development: Machine Learning Applications in High Energy Physics

Machine learning is becoming prevalent in high energy physics, with numerous applications in physics analyses and event reconstruction showing great improvements compared to traditional computing methods. This thesis studies three projects which each propose new avenues for machine learning applications within the high energy physics CMS experiment located at CERN. In the first project, a search for a dark matter signal called “emerging jets” is performed, using graph neural networks to greatly increase sensitivity to the signal’s signature within the data. The result of this dark matter search sets the most stringent exclusion limits to date on theoretical emerging jet models. Motivated by inefficiencies encountered when processing the emerging jet graph neural network at Fermi National Accelerator Laboratory’s computing centers, the second project re-optimizes the computing centers for machine learning inference. This re-optimization uses NVIDIA Triton Inference Servers to process users’ analysis code heterogeneously, therefore achieving high processing throughput and decreasing user time-to-insight. The last project focuses on an upgrade to the CMS experiment’s real-time event selection system which improves physics object reconstruction under harsh processing conditions. A boosted decision tree is used to quickly and efficiently quantify a reconstructed particle’s “track quality” in order to remove particle tracks reconstructed erroneously. In summary, this thesis will not only present examples of how high energy physics can greatly benefit by leveraging machine learning techniques for physics analysis and reconstruction, but will also provide guidance on how the field can prepare for the inevitable increase in machine learning applications.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2024

Quantitative Insight to Fission Gas Pores Distribution in Irradiated Annular U-10Zr Metallic Fuel Using Machine Learning

Metallic fuels, particularly U-10Zr and its performance in reactor irradiation conditions, have been thoroughly investigated and are a promising candidate for next-generation sodium-cooled fast spectrum nuclear reactors. Irradiation in reactors can lead to the formation of fission gas and increased pore formation which can significantly impact fuel performance. Due to the large number of pores and various phases formed in metallic fuel during irradiation, a quantitative description of fission gas pores as a function of irradiation conditions is not yet available, undermining the fidelity of fuel performance modeling to support fuel qualification. It has been difficult to clearly detect pore boundaries and distinguish matrix phases from fission gas pores using optical microscopy by using simple threshold methods working with low magnification images. The pre-trained deep learning model for fission gas pore detection was applied to ~10,260 high magnification scanning electron microscopy images. The model increased the accuracy of fission gas pore segmentation to obtain statistical features, which cannot be processed manually. A pre-trained decision tree model was used to classify pores as isolated or connected pores, providing new insight into the correlation between the movement of lanthanides, solid fission products, and the radial temperature gradient developed in fuel irradiation conditions. This paper emphasizes the potential that artificial intelligence-based machine learning models have to accelerate qualification and support nuclear fuel development.

36 MATERIALS SCIENCE↗
DOE OSTI2024

MicroBooNE investigations on the photon interpretation of the MiniBooNE low energy excess

The MicroBooNE experiment is a liquid argon time projection chamber with 85-ton active volume at Fermilab, operated from 2015 to 2020 to collect neutrino data from Fermilab's Booster Neutrino Beam. One of MicroBooNE's physics goals is to investigate possible explanations of the low-energy excess observed by the MiniBooNE experiment in $\nu_{\mu}\rightarrow \nu_{e}$ neutrino oscillation measurements. MicroBooNE has performed searches to test hypothetical interpretations of the MiniBooNE low-energy excess, including the underestimation of the photon background or instrinic $\nu_{e}$ background. This thesis presents MicroBooNE's searches for two neutral current (NC) single-photon production processes that contribute to the photon background of the MiniBooNE measurement: NC $\Delta$ resonance production followed by $\Delta$ radiative decay: $\Delta \rightarrow N\gamma$, and NC coherent single-photon production. Both searches take advantage of boosted decision trees to yield efficient background rejection, and a high-statistic NC $\pi^0$ measurement to constrain dominant background, and make use of MicroBooNE's first three years of data. The NC $\Delta \rightarrow N\gamma$ measurement yielded a bound on the $\Delta$ radiative decay process at 2.3 times the predicted nominal rate at 90\% confidence level(C.L.), disfavoring a candidate photon interpretation of the MiniBooNE low-energy excess as a factor of 3.18 times the nominal NC Δ radiative decay rate at the 94.8\% C.L. The NC coherent single photon measurement leads to the world's first experimental limit on the cross-section of this process below 1 GeV, of $1.49 \times 10^{-41} \text{cm}^2$ at 90\% C.L., corresponding to 24.0 times the nominal prediction.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2022

Understanding Fission Gas Bubble Distribution and Zirconium Redistribution in Neutron-irradiated U-Zr Metallic Fuel Using Machine Learning

U-10wt.% Zr (U-10Zr) based metallic fuel is the leading candidate for next-generation sodium cooled fast reactor in United States. Currently, Idaho National Laboratory (INL) has been the leading national laboratory for research, development, and demonstration (RD&D) on metallic fuel. Advanced post-irradiation characterization will help to understand fuel microstructure and property change during irradiation, benefiting fuel qualification for commercial application. Characterization capabilities ranging from sub-nanometer to micrometer, such as scanning electron microscopy (SEM), focused ion beam (FIB) sampling, transmission electron microscopy (TEM) characterization, and local thermal conductivity microscopy (TCM), have been utilized recently on irradiated U-10Zr fuel samples to gain a better understanding of nuclear fuel microstructure and property evolution inside a reactor. The FIB/SEM coupled with energy dispersive X-ray spectroscopy (EDS) can capture the essential information to achieve better understanding of fuel behaviors. Inside a nuclear reactor, the phase and microstructure of U-10Zr is constantly changing under neutron bombardment. For example, the gaseous fission product atoms have a limited solubility inside fuel matrix and tend to precipitate out in bubble form, which not only contribute to fuel thermal conductivity degradation but also provide a shortcut for movement of fission products, i.e. lanthanides. The resultant deposition of lanthanides at the cladding inner surface will potentially trigger a chemical reaction/interaction between nuclear fuel and cladding at reactor operational conditions, threatening fuel integrity and safety. FIB/SEM coupled with EDS can provide the fission bubble information as well as probe into phase separation or Zr redistribution, which is fundamental to predict the fuel performance. With high velocity image data generating method, such as FIB/SEM, an automatic way to extract the microstructural information quantitively can better serve the needs from post irradiation characterization. A trained machine learning model, named Decision Tree, is employed to generate a bubble classifier and to categorize bubbles into three categories: isolated bubble, connected without lanthanides, and connected with lanthanides bubbles[3]. This work presents a showcase of this approach on six regions of a fuel cross-section along the radial temperature gradient. We obtained distributions of bubble categories and porosity rates along the six regions. Moreover, a secondary phase U-Zr2 was determined and found on regions 5 and 6. The secondary phase fraction was increasing from 15.61% in region 5 to 34.79% in region 6 based on this approach . This quantitative data offers insights into the lanthanide migration and potentially thermal conductivity degradation. This information from machine learning will be fed into fuel design code for better prediction of fuel performance.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
DOE OSTI2025

Multi-channel, multi-template event reconstruction for SuperCDMS data using machine learning

SuperCDMS SNOLAB uses kilogram-scale germanium and silicon detectors to search for dark matter. Each detector has Transition Edge Sensors (TESs) patterned on the top and bottom faces of a large crystal substrate, with the TESs electrically grouped into six phonon readout channels per face. Noise correlations are expected among a detector's readout channels, in part because the channels and their readout electronics are located in close proximity to one another. Moreover, owing to the large size of the detectors, energy deposits can produce vastly different phonon propagation patterns depending on their location in the substrate, resulting in a strong position dependence in the readout-channel pulse shapes. Both of these effects can degrade the energy resolution and consequently diminish the dark matter search sensitivity of the experiment if not accounted for properly. We present a new algorithm for pulse reconstruction, mathematically formulated to take into account correlated noise and pulse shape variations. This new algorithm fits N readout channels with a superposition of M pulse templates simultaneously - hence termed the N$\times$M filter. We describe a method to derive the pulse templates using principal component analysis (PCA) and to extract energy and position information using a gradient boosted decision tree (GBDT). We show that these new N$\times$M and GBDT analysis tools can reduce the impact from correlated noise sources while improving the reconstructed energy resolution for simulated mono-energetic events by more than a factor of three and for the 71Ge K-shell electron-capture peak recoils measured in a previous version of SuperCDMS called CDMSlite to $<$ 50 eV from the previously published value of $\sim$100 eV. These results lay the groundwork for position reconstruction in SuperCDMS with the N$\times$M outputs.

Albakry, M. F. [British Columbia U.; TRIUMF]↗
DOE OSTI2025

Searching for Neutrino Tridents in the NOvA Near Detector

This dissertation presents a search for neutrino trident production in the NOvA near detector through the coherent ``dimuon" channel: $\nu_\mu +\hspace{1pt}\text{X} \rightarrow \nu_\mu + \mu^- + \mu^+ +\hspace{1pt}\text{X}$. Trident production is a rare, purely electroweak process with sensitivity to physics beyond the Standard Model. The theoretical background, motivation for studying the process, and previous experimental measurements are reviewed. The analysis uses data collected by the NOvA near detector (ND) from Fermilab's Neutrinos at the Main Injector (NuMI) beam between November 2014 and February 2024, corresponding to an exposure of $25.5\times 10^{20}$ protons on target. The ND is a segmented tracking calorimeter located 800~m from the beam target, receiving neutrinos with a mean energy of 2~GeV. A multi-pass background reduction strategy is implemented, including the development of a novel dimuon-specific tracking technique. Trident candidates are identified using a boost ed decision tree classifier trained on simulated signal and background events. Limited background Monte Carlo statistics necessitate the use of functional fits to sideband data, which are extrapolated to estimate backgrounds in the signal region. The unblinded data contain 9 trident-like events, with an estimated background of 5.66 $\pm$ 5.15 events. This yields a best fit estimate of 3.34 tridents compared to the Standard Model prediction of 4.66. A profiled Feldman-Cousins method is used to determine a 90\% confidence interval of [0,9.1] on the number of signal events, corresponding to an upper limit of 1.95$\times$ the Standard Model prediction. This result represents the lowest energy search for trident events to date, and the first experimental contribution to the process in 27 years.

Bowles, Reed Scott [Indiana U.]↗
DOE OSTI2025

Baryon Number Violation Search

Understanding the fundamental forces and symmetries of nature has long been a central goal of particle physics. While the Standard Model (SM) provides a successful framework, it does not guarantee the conservation of baryon number B or lepton number L, thus motivating searches for their violation. Proton decay, a fundamental process violating B, has been at the forefront of experimental searches for decades.The discovery of the weak neutral current in 1973 unified the electromagnetic and weak forces and inspired the creation of Grand Unified Theories (GUTs) that also unify the strong force. In 1974, the first-ever GUT, proposed by Georgi and Glashow, naturally predicted proton decay. This prediction led to an experimental push to validate these theories, and a large underground detector boom was born. Initially designed for proton decay searches, these detectors later proved invaluable to neutrino physics.Although no evidence for proton decay has yet been observed, next-generation large detectors, such as the Deep Underground Neutrino Experiment (DUNE), offer the opportunity to improve on current experimental limits. Utilizing its Liquid Argon Time Projection Chamber (LArTPC) technology, DUNE is positioned to probe rare processes such as proton decay with increased sensitivity.This thesis presents a sensitivity study for the dominant proton decay mode predicted by Supersymmetric GUTs, p → K+ν, utilizing machine learning approaches. Two methods are explored in this thesis: a Boosted Decision Tree (BDT) analysis and a Graphical Neural Network (GNN) analysis with NuGraph. A lifetime limit of 5.36 ± 0.69 × 1033 years for 400 kt-yrs is found using the BDT, while the GNN achieves a lifetime limit of 6.19±1.26×1033 years for 400-kt-yrs. The NuGraph result offers better sensitivity compared to the current limit set by Super-Kamiokande of 5.90 × 1033 while the BDT result offers a slightly lower sensitivity.Additionally, this thesis discusses cross-section work, a first-ever foray into proton decay and atmospheric neutrinos in a vertical drift (VD) DUNE detector, and extensive hardware contributions to the DUNE Far Detector (FD) 1 Module-0, ProtoDUNE-2, which serves as a testbed for the final detector design and installation.

Stokes, Tyler D. [Louisiana State U.] (ORCID:00000↗
DOE OSTI2026

CDRL: Certification-Driven Reinforcement Learning for Neutrino Flavor Model Discovery

Many scientific discovery problems require searching combinatorial hypothesis spaces under complex domain constraints. Reinforcement learning (RL) offers a promising approach, but existing methods rely on scalar rewards that provide limited information about why candidate solutions fail, leading agents to repeatedly explore invalid regions. We introduce Certification-Driven Reinforcement Learning (CDRL), a framework that leverages structured feedback from symbolic reasoning tools. When a candidate violates domain constraints, these tools produce certificates identifying the actions responsible for failure. CDRL converts these certificates into reusable constraints that eliminate classes of invalid solutions and guide exploration toward valid regions. We evaluate CDRL on neutrino flavor model discovery in theoretical particle physics, where the hypothesis space exceeds $10^{26}$ possible models, and compare it with the state-of-the-art RL approach previously used for this task. Across three theory spaces, CDRL achieves up to 1.95$\times$ higher valid model rates and up to 6.33$\times$ higher neutrino model rates while evaluating up to 4$\times$ fewer candidates. We further extract 40 interpretable rules from search trajectories using a post-hoc decision-tree framework and show that reusing them as soft constraints yields gains of up to 2$\times$ in valid model rates and 3$\times$ in neutrino model discovery across all three theory spaces. These results suggest that CDRL uncovers reusable structure in combinatorial search spaces and provides a general framework for scientific model discovery.

Jha, Piyush [Georgia Tech., Atlanta; Georgia Tech]↗
NASA NTRS2010

Systems Maintenance Automated Repair Tasks (SMART)

SMART is a uniform automated discrepancy analysis and repair-authoring platform that improves technical accuracy and timely delivery of repair procedures for a given discrepancy (see figure a). SMART will minimize data errors, create uniform repair processes, and enhance the existing knowledge base of engineering repair processes. This innovation is the first tool developed that links the hardware specification requirements with the actual repair methods, sequences, and required equipment. SMART is flexibly designed to be useable by multiple engineering groups requiring decision analysis, and by any work authorization and disposition platform (see figure b). The organizational logic creates the link between specification requirements of the hardware, and specific procedures required to repair discrepancies. The first segment in the SMART process uses a decision analysis tree to define all the permutations between component/ subcomponent/discrepancy/repair on the hardware. The second segment uses a repair matrix to define what the steps and sequences are for any repair defined in the decision tree. This segment also allows for the selection of specific steps from multivariable steps. SMART will also be able to interface with outside databases and to store information from them to be inserted into the repair-procedure document. Some of the steps will be identified as optional, and would only be used based on the location and the current configuration of the hardware. The output from this analysis would be sent to a work authoring system in the form of a predefined sequence of steps containing required actions, tools, parts, materials, certifications, and specific requirements controlling quality, functional requirements, and limitations.

Schuh, Joseph↗
DOE OSTI2023

A Continuous Action Space Tree search for INverse desiGn (CASTING) framework for materials discovery

Abstract Material properties share an intrinsic relationship with their structural attributes, making inverse design approaches crucial for discovering new materials with desired functionalities. Reinforcement Learning (RL) approaches are emerging as powerful inverse design tools, often functioning in discrete action spaces. This constrains their application in materials design problems, which involve continuous search spaces. Here, we introduce an RL-based framework CASTING (Continuous Action Space Tree Search for inverse design), that employs a decision tree-based Monte Carlo Tree Search (MCTS) algorithm with continuous space adaptation through modified policies and sampling. Using representative examples like Silver (Ag) for metals, Carbon (C) for covalent systems, and multicomponent systems such as graphane, boron nitride, and complex correlated oxides, we showcase its accuracy, convergence speed, and scalability in materials discovery and design. Furthermore, with the inverse design of super-hard Carbon phases, we demonstrate CASTING’s utility in discovering metastable phases tailored to user-defined target properties and preferences.

36 MATERIALS SCIENCE↗
DOE OSTI2020

Perovskite neural trees

Trees are used by animals, humans and machines to classify information and make decisions. Natural tree structures displayed by synapses of the brain involves potentiation and depression capable of branching and is essential for survival and learning. Demonstration of such features in synthetic matter is challenging due to the need to host a complex energy landscape capable of learning, memory and electrical interrogation. We report experimental realization of tree-like conductance states at room temperature in strongly correlated perovskite nickelates by modulating proton distribution under high speed electric pulses. This demonstration represents physical realization of ultrametric trees, a concept from number theory applied to the study of spin glasses in physics that inspired early neural network theory dating almost forty years ago. We apply the tree-like memory features in spiking neural networks to demonstrate high fidelity object recognition, and in future can open new directions for neuromorphic computing and artificial intelligence.

36 MATERIALS SCIENCE↗
DOE OSTI2023

Predictive understanding of the surface tension and velocity of sound in ionic liquids using machine learning

Knowledge of the physical properties of ionic liquids (ILs), such as the surface tension and speed of sound, is important for both industrial and research applications. Unfortunately, technical challenges and costs limit exhaustive experimental screening efforts of ILs for these critical properties. Previous work has demonstrated that the use of quantum-mechanics-based thermochemical property prediction tools, such as the conductor-like screening model for real solvents, when combined with machine learning (ML) approaches, may provide an alternative pathway to guide the rapid screening and design of ILs for desired physiochemical properties. However, the question of which machine-learning approaches are most appropriate remains. In the present study, we examine how different ML architectures, ranging from tree-based approaches to feed-forward artificial neural networks, perform in generating nonlinear multivariate quantitative structure–property relationship models for the prediction of the temperature- and pressure-dependent surface tension of and speed of sound in ILs over a wide range of surface tensions (16.9–76.2 mN/m) and speeds of sound (1009.7–1992 m/s). The ML models are further interrogated using the powerful interpretation method, shapley additive explanations. We find that several different ML models provide high accuracy, according to traditional statistical metrics. The decision tree-based approaches appear to be the most accurate and precise, with extreme gradient-boosting trees and gradient-boosting trees being the best performers. However, our results also indicate that the promise of using machine-learning to gain deep insights into the underlying physics driving structure–property relationships in ILs may still be somewhat premature.

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

Accelerating Random Forest Classification on GPU and FPGA

Random Forests (RFs) are a commonly used machine learning method for classification and regression tasks spanning a variety of application domains, including bioinformatics, business analytics, and software optimization. While prior work has focused primarily on improving performance of the training of RFs, many applications, such as malware identification, cancer prediction, and banking fraud detection, require fast RF classification. In this work, we accelerate RF classification on GPU and FPGA. In order to provide efficient support for large datasets, we propose a hierarchical memory layout suitable to the GPU/FPGA memory hierarchy. We design three RF classification code variants based on that layout, and we investigate GPU- and FPGA-specific considerations for these kernels. Our experimental evaluation, performed on an Nvidia Xp GPU and on a Xilinx Alveo U250 FPGA accelerator card using publicly available datasets on the scale of millions of samples and tens of features, covers various aspects. First, we evaluate the performance benefits of our hierarchical data structure over the standard compressed sparse row (CSR) format. Second, we compare our GPU implementation with cuML, a machine learning library targeting Nvidia GPUs. Third, we explore the performance/accuracy tradeoff resulting from the use of different tree depths in the RF. Finally, we perform a comparative performance analysis of our GPU and FPGA implementations. Our evaluation shows that for high accuracy targets, our GPU implementation yields 5-9x speedup over CSR, and up to a 2x speedup over cuML.

FPGA, Xilinx FPGA, GPU, Random Forest classificati↗
DOE OSTI2021

Mesoscopic Modeling and Rapid Simulation of Incremental Changes in Epidemic Scenarios on GPUs

In simulation-based studies and analyses of epidemics, a major challenge lies in resolving the conflict between fidelity of models and the speed of their simulation. Another related challenge arises in dealing with the large number of what–if scenarios that need to be explored. Here, we describe new computational methods that together provide an approach to dealing with both challenges. A mesoscopic modeling approach is described that strikes a middle ground between macroscopic models based on coupled differential equations and microscopic models built on fine-grained behaviors at the individual entity level. The mesoscopic approach offers the ability to incorporate complex compositions of multiple layers of dynamics even while retaining the potential for aggregate behaviors at varying levels. It also is an excellent match to the accelerator-based architectures of modern computing platforms in which graphical processing units (GPUs) can be exploited for fast simulation via the parallel execution mode of single instruction multiple thread (SIMT). The challenge of simulating a large number of scenarios is addressed via a method of sharing model state and computation across a tree of what–if scenarios that are localized, incremental changes to a large base simulation. A combination of the mesoscopic modeling approach and the incremental what–if scenario tree evaluation has been implemented in the software on modern GPUs. Synthetic simulation scenarios are presented to demonstrate the computational characteristics of our approach. Results from the experiments with large population data, including USA, UK, and India, illustrate the modeling methodology and computational performance on thousands of synthetically generated what–if scenarios. Execution of our implementation scaled to 8192 GPUs of supercomputing platforms demonstrates the ability to rapidly evaluate what–if scenarios several orders of magnitude faster than the conventional methods.

97 MATHEMATICS AND COMPUTING↗