A novel machine learning based lumping approach for the reduction of large kinetic mechanisms for plasma-assisted combustion applications
Not Available
SEARCH · Search NASA
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.
Not Available
Resource allocation in High Performance Computing (HPC) environments presents a complex and multifaceted challenge for job scheduling algorithms. Beyond the efficient allocation of system resources, schedulers must account for and optimize multiple performance metrics, including job wait time and system throughput. Traditional heuristic-based scheduling algorithms increasingly struggle and lack the efficiency needed to meet the demands and address the complexity and scale of modern HPC systems. Consequently, recent research efforts have focused on leveraging advancements in Artificial Intelligence (AI) and Deep Learning (DL), particularly Reinforcement Learning (RL), to develop more adaptable and intelligent scheduling strategies. Previous RL-based scheduling approaches have explored a range of algorithms, from Deep Q-Networks (DQN) to Proximal Policy Optimization (PPO), and more recently, hybrid methods that integrate Graph Neural Networks (GNNs) with RL techniques. However, a common limitation across these methods is their reliance on relatively small datasets, with few methods being evaluated using large-scale, multi-million-job trace datasets representative of real-world HPC workloads. Moreover, existing RL schedulers face scalability issues due to centralized policy updates, which hinder training efficiency and performance when applied to large datasets. This study introduces a novel RL-based scheduler utilizing Decentralized Distributed Proximal Policy Optimization (DD-PPO) algorithm, which supports large-scale distributed training across multiple workers without requiring parameter synchronization at every step. By eliminating reliance on centralized updates to a shared policy, the DD-PPO scheduler enhances scalability, training efficiency, and sample utilization. Experimental validation using a large real-world dataset containing over 11.5 million job traces collected from petascale HPC systems over six years assesses the influence of dataset scale on training effectiveness and compares DD-PPO performance to traditional and advanced scheduling approaches. The experimental results demonstrate improved scheduling performance in comparison to both heuristic-based schedulers and existing RL-based scheduling algorithms.
Plant bioengineering has generated tens of thousands of genotype-to-phenotype relationships, but this knowledge remains fragmented across narrative literature and difficult to use computationally. Inconsistent descriptions of DNA constructs, host species, and traits, including variable species names, omitted regulatory elements, and inconsistent gene symbols, impede data reuse, comparative analysis, and design-build-test-learn cycles. Here, we present the Plant Bioengineering Atlas, a literature-mined, ontology-grounded knowledge base assembled using an artificial intelligence (AI)-aided extraction pipeline. A large language model parsed open-access primary research articles to generate structured, provenance-anchored records of engineered genes, modification types, promoter-gene-terminator constructs, host species, target traits, and reported phenotypes, with every record traceable to its source. The current release contains 14,358 curated records encompassing 6,998 distinct genes across 436 plant species from 6,452 papers published between 2000 and 2026. Corpus analysis reveals that experiments are concentrated in a small group of model and crop species, disease and pathogen resistance is the most frequently engineered trait class, and constitutive regulatory parts (particularly the CaMV 35S promoter and NOS terminator) remain pervasive. Two in five records omit one or both flanking regulatory elements (i.e., promoter and terminator), while only 23.4% describe cassettes in which both elements resolve to named part classes, exposing a systematic reproducibility gap. We organize these data into a knowledge graph linking genes, constructs, species, and traits; provide access through an interactive web portal; and propose an AI-compatible documentation standard for AI-ready reporting. The Plant Bioengineering Atlas provides a foundation for data-driven hypothesis generation and AI-aided plant biodesign.
Ensuring power grid resiliency, forecasting climate conditions, and optimization of transportation infrastructure are some of the many application areas where data is collected in both space and time. Spatiotemporal modeling is about modeling those patterns for forecasting future trends and carrying out critical decision-making by leveraging machine learning/deep learning. Once trained offline, field deployment of trained models for near real-time inference could be challenging because performance can vary significantly depending on the environment, available compute resources and tolerance to ambiguity in results. Users deploying spatiotemporal models for solving complex problems can benefit from analytical studies considering a plethora of system adaptations to understand the associated performance-quality trade-offs. To facilitate the co-design of next-generation hardware architectures for field deployment of trained models, it is critical to characterize the workloads of these deep learning (DL) applications during inference and assess their computational patterns at different levels of the execution stack. In this paper, we develop several variants of deep learning applications that use spatiotemporal data from dynamical systems. We study the associated computational patterns for inference workloads at different levels, considering relevant models (Long short-term Memory, Convolutional Neural Network and Spatio-Temporal Graph Convolution Network), DL frameworks (Tensorflow and PyTorch), precision (FP16, FP32, AMP, INT16 and INT8), inference runtime (ONNX and AI Template), post-training quantization (TensorRT) and platforms (Nvidia DGX A100 and Sambanova SN10 RDU). Overall, our findings indicate that although there is potential in mixed-precision models and post-training quantization for spatiotemporal modeling, extracting efficiency from contemporary GPU systems might be challenging. Instead, co-designing custom accelerators by leveraging optimized High Level Synthesis frameworks (such as SODA High-Level Synthesizer for customized FPGA/ASIC targets) can make workload-specific adjustments to enhance the efficiency.
The Computer Science Research Institute (CSRI) brings university faculty and students to Sandia National Laboratories for focused collaborative research on Department of Energy (DOE) computer and computational science problems. The institute provides an opportunity for university researches to learn about problems in computer and computational science at DOE laboratories, and help transfer results of their research to programs at the labs. Some specific CSRI research interest areas are: scalable solvers, optimization, algebraic preconditioners, graph-based, discrete, and combinatorial algorithms, uncertainty estimation, validation and verification methods, mesh generation, dynamic load-balancing, virus and other malicious-code defense, visualization, scalable cluster computers, beyond Moore’s Law computing, exascale computing tools and application design, reduced order and multiscale modeling, parallel input/output, and theoretical computer science. The CSRI Summer Program is organized by CSRI and includes a weekly seminar series and the publication of a summer proceedings.
The Computer Science Research Institute (CSRI) brings university faculty and students to Sandia National Laboratories for focused collaborative research on Department of Energy (DOE) computer and computational science problems. The institute provides an opportunity for university researches to learn about problems in computer and computational science at DOE laboratories, and help transfer results of their research to programs at the labs. Some specific CSRI research interest areas are: scalable solvers, optimization, algebraic preconditioners, graph-based, discrete, and combinatorial algorithms, uncertainty estimation, validation and verification methods, mesh generation, dynamic load-balancing, virus and other malicious-code defense, visualization, scalable cluster computers, beyond Moore’s Law computing, exascale computing tools and application design, reduced order and multiscale modeling, parallel input/output, and theoretical computer science. The CSRI Summer Program is organized by CSRI and includes a weekly seminar series and the publication of a summer proceedings.
The objective of this project is to test the Algorithm to Architecture Mapping Model (ATAMM) firing rules using the Mentat run-time system and the Mentat Programming Language (MPL). A special version of Mentat, Mentat/A (Mentat/ATAMM) was constructed. This required changes to: (1) modify the run-time system to control queue length and inhibit actor firing until required data tokens are available and space is available in the input queues of all of the direct descendent actors; (2) disallow the specification of persistent object classes in the MPL; and (3) permit only decision free graphs in the MPL. We were successful in implementing the spirit of the plan, although some goals changed as we came to better understand the problem. We report on what we accomplished and the lessons we learned. The Mentat/A run-time system is discussed, and we briefly present the compiler. We present results for three applications and conclude with a summary and some observations. Appendix A contains a list of technical reports and published papers partially supported by the grant. Appendix B contains listings for the three applications.
Anomaly detection is a common machine learning (ML) task with growing importance in the fields of imaging, quality assurance, and multiple security related disciplines. Anomaly detection is more difficult than traditional machine learning methods due to the inherent unlabeled nature of the datasets. Existing anomaly detection architectures commonly face challenges with explainability, retaining information related to the relational structure of the data, and false positive rates. Hyperspectral Imaging Anomaly Detection (HSI) is a statistical model that employs vertex and edge weighted graphs to preserve the data’s relationships on different topographical scales. The model is able to generalize from anomaly detection in 2D images to novel datasets related to cyber-security. Furthermore, the use of multi-spectral and other filtering methods results in fewer false positives and increases the explainability of model predictions. When applying HSI to cyber-security datasets, we are able to successfully detect malicious activity with a relatively high degree of accuracy.
We study the applicability of spiking neural networks and neuromorphic hardware for solving general optimization problems without the use of adaptive training or learning algorithms. We leverage the dynamics of Hopfield networks and spin-glass systems to construct a fully connected spiking neural system to generate synchronous spike responses indicative of the underlying community structure in an undirected, unweighted graph. Mapping this fully connected system to current generation neuromorphic hardware is done by embedding sparse tree graphs to generate only the leading-order spiking dynamics. Here, we demonstrate that for a chosen set of benchmark graphs, the spike responses generated on a current generation neuromorphic processor can improve the stability of graph partitions and non-overlapping communities can be identified even with the loss of higher-order spiking behavior if the graphs are sufficiently dense. For sparse graphs, the loss of higher-order spiking behavior improves the stability of certain graph partitions but does not retrieve the known community memberships.
Stochastic physical systems across the sciences that have very high-dimensional state spaces, with a large number of fast degrees of freedom that force direct simulators to proceed by integration steps that are orders of magnitude smaller than events of interests (e.g., particle collisions). Examples range from molecular motion to dynamics of large populations of cells. A grand challenge in the simulation and understanding of such systems is the systematic construction of accurate, interpretable, reduced models, enabling faster simulations, revealing fundamental properties of the dynamics, and predicting phenomena of interest that the original simulator could not reached with sufficient accuracy or within a given computational budget. In this projected we developed novel statistical estimation/machine learning techniques for analyzing and building empirical reduced models for important families of high-dimensional stochastic systems, in particular: - we developed techniques for estimating interaction kernels in interacting particle- and agent-based systems, which are ubiquitous in Physics, Biology and many other sciences, given observed trajectories of the system; - we developed techniques for nonlinear model reduction for high-dimensional stochastic systems that have a small number of unknown, nonlinear slow variables, and a large number of fast modes, that are possibly of large magnitude, given observed short trajectories of the system in the form of bursts of trajectories from different initial conditions; - we developed novel techniques for estimating linear dynamical systems on graphs when both the dynamics and the underlying graph are unknown, and we have a sparse set of space-time observations; - we considered the problem of estimating an unknown nonlinear observation function of a standard process (e.g. Brownian motion), so that we can recognized if an observed dynamics is "just" a nonlinear version of a known dynamics; we also developed benchmarks for learning algorithms aimed at learning and classifying diffusion processes.
The ability of machine learning (ML) classification models to resist small, targeted input perturbations—known as adversarial attacks—is a key measure of their safety and reliability. We show that floating-point non associativity (FPNA) coupled with asynchronous parallel programming on GPUs is sufficient to result in misclassification, without any perturbation to the input. Additionally, we show that this misclassification is particularly significant for inputs close to the decision boundary and that standard adversarial robustness results may be overestimated up to 4.6 when not considering machine-level details. We first study a linear classifier, before focusing on standard Graph Neural Network (GNN) architectures and datasets used in robustness assessments. We develop a novel black-box attack using Bayesian optimization to discover external workloads that can change the instruction scheduling which bias the output of reductions on GPUs and reliably lead to misclassification. Motivated by these results, we present a new learnable permutation (LP) gradient-based approach to learning floating-point operation orderings that lead to misclassifications. The LP approach provides a worst-case estimate in a computationally efficient manner, avoiding the need to run identical experiments tens of thousands of times over a potentially large set of possible GPU states or architectures. Finally, using instrumentation-based testing, we investigate parallel reduction ordering across different GPU architectures under external background workloads, when utilizing multi-GPU virtualization, and when applying power capping. Our results demonstrate that parallel reduction ordering varies significantly across architectures under the first two conditions, substantially increasing the search space required to fully test the effects of this parallel scheduler-based vulnerability. These results and the methods developed here can help to include machine-level considerations into adversarial robustness assessments, which can make a difference in safety and mission critical applications.
The accurate prediction of molecular properties is important for the rational design and the advancement of green chemistry and sustainable materials research. However, the predictive power of traditional computational chemistry methods is limited due to computational restrictions. Here, in this study, we examine an alternative approach to the accurate prediction of properties of organic compounds: natural language processing (NLP)-based molecular embedding. Using viscosity, partition coefficient (log P), and enthalpy of vaporization as test properties through a survey of comprehensive datasets comprising 5695 data points for viscosity, 25 870 data points for log P, and 2296 data points for enthalpy of vaporization. These are important properties for the design of greener, safer, and sustainable chemical processes. Models were trained using NLP methods such as Mol2vec and fine-tuned ChemBERTa, and results were compared with traditional input featurization techniques such as Morgan fingerprints and quantum chemistry derived sigma profiles and DFT features. Among the various machine learning models, Mol2vec demonstrated superior predictive capabilities, achieving the highest correlation coefficient (R 2 = 0.945) and lowest RMSE (0.106 mPa s) for viscosity, as well as high accuracy for log P and enthalpy of vaporization predictions. These findings establish the Mol2vec featurization technique, graph-convolutional neural networks (GCNN), and fine-tuned ChemBERTa model as powerful tools for predictive modeling of organic compounds properties, offering a significant improvement over previously used featurization techniques and opening up strategies for very-high-throughput computational screening. Finally, we integrated ML models with hybrid language-model-based generative adversarial networks (LM-GAN) to generate novel molecular sequences with desirable properties for different research applications. The ability to computationally design solvents with lower viscosity, lower log P, and lower enthalpy of vaporization offers a data-driven route to accelerating the discovery of sustainable alternatives to traditionally toxic solvents.
Network biology is an interdisciplinary field bridging computational and biological sciences that has proved pivotal in advancing the understanding of cellular functions and diseases across biological systems and scales. Although the field has been around for two decades, it remains nascent. It has witnessed rapid evolution, accompanied by emerging challenges. These stem from various factors, notably the growing complexity and volume of data together with the increased diversity of data types describing different tiers of biological organization. We discuss prevailing research directions in network biology, focusing on molecular/cellular networks but also on other biological network types such as biomedical knowledge graphs, patient similarity networks, brain networks, and social/contact networks relevant to disease spread. In more detail, we highlight areas of inference and comparison of biological networks, multimodal data integration and heterogeneous networks, higher-order network analysis, machine learning on networks, and network-based personalized medicine. Following the overview of recent breakthroughs across these five areas, we offer a perspective on future directions of network biology. Additionally, we discuss scientific communities, educational initiatives, and the importance of fostering diversity within the field. This article establishes a roadmap for an immediate and long-term vision for network biology.
Purpose: Traditionally, breast cancer metastasis is described as a process wherein cancer cells spread from the breast to multiple organ systems via hematogenous and lymphatic routes. Mapping organ specific patterns of cancer spread over time is essential to understanding metastatic progression. In order to better predict sites of metastases, here we demonstrate modeling of the patterned migration of metastasis. Methods: We reviewed the clinical history of 453 breast cancer patients from Memorial Sloan Kettering Cancer Center who were non-metastatic at diagnosis but developed metastasis over time. We used the variables of organ site of metastases as well as time to create a Markov chain model of metastasis. We illustrate the probabilities of metastasis occurring at a given anatomic site together with the probability of spread to additional sites. Results: Based on the clinical histories of 453 breast cancer patients who developed metastasis, we have learned (i) how to create the Markov transition matrix governing the probabilities of cancer progression from site to site; (ii) how to create a systemic network diagram governing disease progression modeled as a random walk on a directed graph; (iii) how to classify metastatic sites as ‘sponges’ that tend to only receive cancer cells or ‘spreaders’ that receive and release them; (iv) how to model the time-scales of disease progression as a Weibull probability distribution function; (v) how to perform Monte Carlo simulations of disease progression; and (vi) how to interpret disease progression as an entropy-increasing stochastic process. Conclusion: Based on our modeling, metastatic spread may follow predictable pathways. Mapping metastasis not simply by organ site, but by function as either a ‘spreader’ or ‘sponge’ fundamentally reframes our understanding of metastatic processes. This model serves as a novel platform from which we may integrate the evolving genomic landscape that drives cancer metastasis. PS-OC Trans-Network Project Grant Award for “Data Assimilation and ensemble statistical forecasting methods applied to the MSKCC longitudinal metastatic breast cancer cohort.”.
The objective of this project is to support the Algorithm to Architecture Mapping Model (ATAMM) firing rules using the Mentat run-time system and the Mentat Programming Language (MPL). Specifically, this required changes to (1) modify the run-time system to control queue length and inhibit actor firing until required data tokens are available and space is available in the input queues of all of the direct descendent actors, (2) disallow the specification of persistent object classes in the MPL, and (3) permit only decision free graphs in the MPL. We have been successful in implementing the spirit of the plan, although some goals changed as we came to better understand the problem. Here, we report on what we accomplished and the lessons we learned. The Mentat/A runtime system, the compiler, results for three applications, and a summary and some observations are given. Appendix A contains a list of technical reports and published papers partially supported by the grant. Appendix B contains listings for the three applications.
Artificial intelligence (AI) holds immense promise for revolutionizing microscopy, yet its widespread adoption has been hindered by challenges ranging from user inexperience to limited model transferability and difficulties in operationalizing machine learning. This presentation showcases our approach to developing practical autonomy for materials discovery, aiming to accelerate the integration of AI into everyday microscopy workflows. As shown in Fig. 1, I will focus on three key areas: understanding order-disorder transitions, quantifying point defects, and achieving truly device-scale microscopy. First, I will demonstrate the power of multi-modal knowledge graphs for integrating diverse microscopy data. By combining imaging, spectroscopy, and diffraction data, these graphs provide a holistic view of material behavior, capturing the intricate relationships between different modalities [1,2]. I will present a case study on how these models illuminate the structural and chemical changes associated with irradiation in oxide thin films, revealing critical insights for designing materials for extreme environments like spaceflight and nuclear energy. Specifically, I will show how multi-modal analysis clarifies the evolution of order-disorder transitions under irradiation, a key factor influencing material performance in these applications. Next, I will address the challenge of quantifying point defects in 2D materials. We demonstrate the application of computer vision and transfer learning to accurately identify and classify various defect types, such as vacancies and substitutional atoms, and to quantify their concentrations. This information is crucial for understanding and tailoring the properties of 2D materials for applications in electronics, optoelectronics, and catalysis. For example, I will show how our models can characterize the topological distribution of point defects in MXene transition metal carbides, providing valuable insights for optimizing their performance in energy storage and separation science. Finally, I will discuss our progress toward autonomous device-scale microscopy [3,4]. We are fundamentally redesigning electron microscopes around the principles of machine reasoning, enabling automation beyond basic tasks like sample navigation and data acquisition to include sophisticated experimental design. This approach paves the way for truly reproducible and massively scaled analysis campaigns. I will emphasize the importance of autonomous microscopy platforms for high-throughput materials discovery and characterization, facilitating the rapid screening of materials for a broad range of applications and accelerating the development of next-generation technologies.
Drug discovery is a complex and challenging process, requiring the optimization of candidate compounds to identify those with the potential to become safe and effective drugs. Predicting molecular properties is an indispensable step in the drug discovery pipeline. Traditionally, this process is costly and time-intensive, involving multiple rounds of experiments and clinical trials, rendering it impractical for every candidate compound. Deep learning techniques have emerged as a promising approach to drug discovery to reduce the cost and time required to identify novel drugs. However, prevalent research in deep learning models focused on predicting molecular properties has primarily fixated on single-modal models, which utilize a single modality of data, neglecting the potential benefits of combining different data modalities. To overcome this limitation, we introduce MRL-Mol: a deep \textbf{M}ultimodal \textbf{R}epresentation \textbf{L}earning framework for accurate \textbf{Mol}ecular properties prediction. MRL-Mol harnesses three data modalities: sequence, graph, and image, augmenting the depth of comprehension. Leveraging a large-scale unlabeled dataset~($\sim$1M unique molecules), we pretrain MRL-Mol to extract inter- and intra-modal information. Our study demonstrates the superior performance of MRL-Mol in predicting molecular properties across six benchmark datasets, including both classification and regression tasks. Notably, MRL-Mol outperforms other state-of-the-art molecular properties prediction models. These findings suggest that by combining information from multiple data modalities, MRL-Mol can comprehend molecules better than single-modal deep learning models and identify molecular properties with better accuracy.
Traditional methods for identifying “hit” molecules from a large collection of potential drug-like candidates rely on biophysical theory to compute approximations to the Gibbs free energy of the binding interaction between the drug and its protein target. These approaches have a significant limitation in that they require exceptional computing capabilities for even relatively small collections of molecules. Increasingly large and complex state-of-the-art deep learning approaches have gained popularity with the promise to improve the productivity of drug design, notorious for its numerous failures. However, as deep learning models increase in their size and complexity, their acceleration at the hardware level becomes more challenging. Hyperdimensional Computing (HDC) has recently gained attention in the computer hardware community due to its algorithmic simplicity relative to deep learning approaches. The HDC learning paradigm, which represents data with high-dimension binary vectors, allows the use of low-precision binary vector arithmetic to create models of the data that can be learned without the need for the gradient-based optimization required in many conventional machine learning and deep learning methods. This algorithmic simplicity allows for acceleration in hardware that has been previously demonstrated in a range of application areas (computer vision, bioinformatics, mass spectrometery, remote sensing, edge devices, etc.). To the best of our knowledge, our work is the first to consider HDC for the task of fast and efficient screening of modern drug-like compound libraries. We also propose the first HDC graph-based encoding methods for molecular data, demonstrating consistent and substantial improvement over previous work. We compare our approaches to alternative approaches on the well-studied MoleculeNet dataset and the recently proposed LIT-PCBA dataset derived from high quality PubChem assays. We demonstrate our methods on multiple target hardware platforms, including Graphics Processing Units (GPUs) and Field Programmable Gate Arrays (FPGAs), showing at least an order of magnitude improvement in energy efficiency versus even our smallest neural network baseline model with a single hidden layer. Our work thus motivates further investigation into molecular representation learning to develop ultra-efficient pre-screening tools. We make our code publicly available at https://github.com/LLNL/hdbind.