Search NASA⌕ Search

SEARCH · Search NASA

Results for “database for artificial intelligence training”

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 37 records · Page 2

Aero-Engines AI - A Machine-Learning App for Aircraft Engine Concepts Assessment

Effective deployment of machine-learning (ML) models could drive a high level of efficiency in aircraft engine conceptual design. Aero-Engines AI is a user-friendly app that has been created to deploy trained machine-learning (ML) models to assess aircraft engine concepts. It was created using tkinter, a GUI (graphical user interface) module that is built into the standard Python library. Employing tkinter greatly facilitates the sharing of ML application as an executable file which can be run on Windows machines (without the need to have Python or any library installed). The app gets user input for a turbofan design, preprocesses the input data, and deploys trained ML models to predict turbofan thrust specific fuel consumption (TSFC), engine weight, core size, and turbomachinery stage-counts. The ML predictive models were built by employing supervised deep-learning and K-nearest neighbor regression algorithms to study patterns in an existing open-source database of production and research turbofan engines. They were trained, cross-validated, and tested in Keras, an open-source neural networks API (application programming interface) written in Python, with TensorFlow (Google open-source artificial intelligence library) serving as the backend engine. The smooth deployment of these ML models using the app shows that Aero-Engines AI is an easy-touse and a time-saving tool for aircraft engine design-space exploration during the conceptual design stage. Current version of the app focuses on the performance prediction of conventional turbofans. However, the scope of the app can easily be easily expanded to include other engine types (such as turboshaft and hybrid-electric systems) after their ML models are developed. Overall, the use of a machine-learning app for aircraft engine concept assessment represents a promising area of development in aircraft engine conceptual design.

machine learning↗

Earth Independent Medical Operations (EIMO) DATASCOPE Technical Interchange Meeting 21st August 2023: Background and Summary of Discussion

An aspiration for EIMO datascope is to realize artificial intelligence-enhanced solutions for analysis of crew health & performance data and to facilitate clinical decision support for autonomous medical operations. A vision proposed to the meeting participants was that of a “system of systems,” whereby EIMO will utilize AI-supported natural language processing and machine learning techniques to synthesize embedded reference databases and real-time data streams [input vectors] from multiple data sources to continuously and seamlessly assess crew health & performance. Constituent input vectors may include environmental controls, countermeasures data, behavioral data, physiologic wearables, point-of-care laboratory tests, personalized medical records, inventory trade space risk assessments, COTS medical databases, and ground support inputs. An ideal AI capability would possess trained fusion algorithms to cross reference input vectors with medical ‘knowledge’ [cultivated database] to stratify relevant data streams for predictive and actionable capabilities. In addition, EIMO will ideally have a degree of mobility, in that it can be accessed and can push/pull data within and between multiple vehicles/habitats.

Artificial Intelligence↗

Increasing accessibility to deep learning-based analytics for space biology: pretrained models, transfer learning, and analytics platform development

Biological systems react in complex ways to the stressors of spaceflight, and the data capturing these relationships is concomitantly high-dimensional and complex. Deep learning and machine learning approaches are increasingly popular as an analytical approach for space biosciences, due to their ability to model complex relationships in complex data. However, such approaches often require large datasets and extensive computational resources. New approaches that minimize data sizes and computational power needed to leverage machine learning, and resources that make these approaches accessible, are needed to increase accessibility and adoption of machine learning in the space biosciences. Transfer learning, in which a pretrained model of broad utility is trained on a large dataset, and subsequently reused on downstream applications for which data is more limited, is one approach to minimizing data and computational intensity of deep learning applications. This transfer learning approach results in more performant models in high-dimensional, low-sample-size settings such as space biology, as compared to training models on limited data from scratch. This presentation will outline efforts to generate pretrained models for the space biology community, and highlight transfer learning applications modeling microbial antibiotic resistance during spaceflight. Finally, in order to increase accessibility of these models and tools, as well as others, for the broader space biology community, we present a modeling and analysis platform facilitating machine learning applications in space biology. This platform streamlines machine learning training and analysis in a notebook format, facilitates download and use of space biology data from the NASA GeneLab database, and can be utilized on NASA-hosted servers or downloaded and hosted locally. This effort, as part of the AI4LS (Artificial Intelligence for Life in Space) working group, will increase accessibility, feasibility, and performance of machine learning approaches for the space biology community.

Adrienne Hoarfrost↗

Leveraging Artificial Intelligence to Predict Novel Eutectic Alloys

The goal of this project was to train an artificial neural network (ANN) to predict the fractional composition and melting point of eutectic alloys using fundamental atomic properties as inputs. The fundamental properties considered include atomic number, atomic weight, atomic radius, valence electron concentration, electronegativity, and electron affinity. The project involved several phases, starting with data preparation, where phase diagram data was harvested from the ASM International database. Approximately 1300 binary eutectics were collected and cleaned to ensure relevance and accuracy. A regression model was selected for training, utilizing a rectified linear unit as the activation function. Various model configurations were evaluated for predictive accuracy, with validation techniques employed to ensure robustness. The model demonstrated predictive capabilities above random guessing and was able to achieve up to 11% accuracy under certain conditions. An ablative test identified atomic radius and valence electron concentration as critical inputs for model performance. Incorporating the melting point of atomic constituents improved accuracy significantly, although ultimately the model’s predictive capability still fell short of the 80% target. This report details the methodology, results, and implications of the research, contributing to the understanding of employing artificial intelligence to predict the phase transition behavior of eutectic alloys.

36 MATERIALS SCIENCE↗

Comparison of machine learning systems trained to detect Alfvén eigenmodes using the CO 2 interferometer on DIII-D

Abstract A Machine-Learning (ML) based detection scheme that automatically detects Alfvén Eigenmodes (AE) in a labelled DIII-D database is presented here. Controlling AEs is important for the success of planned burning plasma devices such as ITER, since resonant fast ions can drive AEs unstable and degrade the performance of the plasma or damage the first walls of the machine vessel. Artificial Intelligence could be useful for real-time detection and control of AEs in steady-state plasma scenarios by implementing ML-based models into control algorithms that drive actuators for mitigation of AE impacts. Thus, the objective is to compare differences in performance between using two different recurrent neural network systems (Reservoir Computing Network and Long Short Term Memory Network) and two different representations of the C O 2 phase data (simple and crosspower spectrograms). All C O 2 interferometer chords are used to train both models, but only one is processed during each training step. The results from the model and data comparison show higher performance for the RCN model (True Positive Rate = 90% and False Positive Rate = 14%), and that using simple magnitude spectrograms is sufficient to detect AEs. Also, the vertical C O 2 interferometer chord passing near the center is better for ML-based detection of AEs.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Application of Selected Artificial Intelligence Methods in a System Predicting the Microstructure of Compacted Graphite Iron

Intelligent computing tools such as artificial neural network and fuzzy logic are used as predictive modeling tools. The use of these methods, combined with model experimental results, may be an excellent predictive tool, allowing us to forecast the microstructure of the tested cast iron at the level of computer simulation. In this study, the reference training cases collected in one database were used to determine the parameters of the neuro-fuzzy ANFIS model. They mainly include the results of observations and measurements of the content of individual microstructural constituents of the compacted graphite iron, examined as a function of the content of individual alloy additives (molybdenum, nickel and copper introduced in various proportions). The training process of such a fuzzy inference system is done by constantly changing its parameters (parameters of the membership function) and determining new rule conclusions as a result of presenting individual case examples from the training sample. The conducted research has shown the possibility of applying the ANFIS model as a tool to control the chemical composition of compacted graphite iron in the production of castings with high-strength parameters.

36 MATERIALS SCIENCE↗

Application of machine learning and artificial intelligence to extend EFIT equilibrium reconstruction

Recent progress in the application of machine learning (ML)/artificial intelligence (AI) algorithms to improve the Equilibrium Fitting (EFIT) code equilibrium reconstruction for fusion data analysis applications is presented. A device-independent portable core equilibrium solver capable of computing or reconstructing equilibrium for different tokamaks has been created to facilitate adaptation of ML/AI algorithms. A large EFIT database comprising of DIII-D magnetic, motional Stark effect, and kinetic reconstruction data has been generated for developments of EFIT model-order-reduction (MOR) surrogate models to reconstruct approximate equilibrium solutions. Furthermore, a neural-network MOR surrogate model has been successfully trained and tested using the magnetically reconstructed datasets with encouraging results. Other progress includes developments of a Gaussian process Bayesian framework that can adapt its many hyperparameters to improve processing of experimental input data and a 3D perturbed equilibrium database from toroidal full magnetohydrodynamic linear response modeling using the Magnetohydrodynamic Resistive Spectrum - Feedback (MARS-F) code for developments of 3D-MOR surrogate models.

Gaussian process↗

Appendices for Geothermal Exploration Artificial Intelligence Report

The Geothermal Exploration Artificial Intelligence looks to use machine learning to spot geothermal identifiers from land maps. This is done to remotely detect geothermal sites for the purpose of energy uses. Such uses include enhanced geothermal system (EGS) applications, especially regarding finding locations for viable EGS sites. This submission includes the appendices and reports formerly attached to the Geothermal Exploration Artificial Intelligence Quarterly and Final Reports. The appendices below include methodologies, results, and some data regarding what was used to train the Geothermal Exploration AI. The methodology reports explain how specific anomaly detection modes were selected for use with the Geo Exploration AI. This also includes how the detection mode is useful for finding geothermal sites. Some methodology reports also include small amounts of code. Results from these reports explain the accuracy of methods used for the selected sites (Brady Desert Peak and Salton Sea). Data from these detection modes can be found in some of the reports, such as the Mineral Markers Maps, but most of the raw data is included the DOE Database which includes Brady, Desert Peak, and Salton Sea Geothermal Sites.

15 GEOTHERMAL ENERGY↗

GeoThermalCloud for EGS – An Open-source, User-friendly, Scalable AI Workflow for Modeling Enhanced Geothermal Systems

Enhanced Geothermal Systems (EGS) offer a vast potential to expand the use of geothermal energy. Heat is extracted from this engineered system by injecting relatively cold water into subsurface fractures, which are in contact with hot dry rock, and brought back to surface through production wells. Creating EGS requires improving the natural permeability of hot crystalline rocks. In this short conference paper, we present a reproducible workflow for modeling EGS. Our workflow called the GeoThermalCloud (GTC) for EGS, leverages recent advances in machine learning, deep learning, and high-performance computing. This GTC framework is currently being made open-source, user-friendly, and reproducible through python scripts as well as Google Colab/Jupyter Notebooks. This GTC for EGS modeling scripts are made available at https://github.com/SmartTensors/GeoThermalCloud.jl/tree/master/EGS and will constantly be updated to cater for geothermal community. Current GTC framework provides scripts to train deep learning (DL) models for techno-economics and data worth analysis. The Geothermal Design Tool (https://github.com/GeoDesignTool/GeoDT.git), a fast and simplified multi-physics solver, is used to develop a database for training DL models. This short paper provides details on the scripts to curate, process, and train DL models. The scripts can easily be modified to train on databases generated by other popular open-source simulators such as PFLOTRAN, STOMP, TOUGH, and GEOSX or commercial software such as ResFrac and COMSOL.

15 GEOTHERMAL ENERGY↗

Using Rule-Based Models for Weak Supervised Learning: A ChemNet for Transferable Chemical Property Prediction

With access to large datasets, deep neural networks (DNN) have achieved human-level accuracy in image and speech recognition tasks. However, in chemistry, data is inherently small and fragmented. In this work, we develop an approach of integrating rule-based knowledge into the Chemception CNN model via transfer learning techniques. The resulting model, ChemNet, is a transferable and generalizable deep neural network for chemical property prediction that learns in a semi-supervised manner from large unlabeled chemical databases. When ChemNet is further fine-tuned on 3 smaller datasets to predict chemical properties that it was not originally trained on, we show that ChemNet exceeds the accuracy of existing Chemception CNN models and other contemporary DNN models that were trained using conventional supervised learning approaches. These results indicate that pre-training ChemNet on a large diverse chemical database while incorporating chemistry domain knowledge, enables the development of more generalizable deep neural networks for the prediction of novel chemical properties.

Artificial Intelligence, Cheminformatics, Computat↗

A Machine-Learning Approach to Assess Aircraft Engine System Performance

Artificial intelligence (AI)/machine learning, and big data are transforming the global business environment. They have become the most disruptive technologies for organizations to improve workplace efficiency and productivity. This work explored the application of machine learning-based predictive analytics that would enable aircraft engine designers to estimate engine system performance quickly during the conceptual design stage. Supervised machine-learning algorithm was employed to study patterns in an existing database of production and research turbofan engines, and built predictive analytics for use in predicting system performance of new turbofan designs. Specifically, the author developed deep-learning analytics to predict turbofan system weight, using turbofan design parameters as the input. The predictive analytics were trained and deployed in Keras, an open-source neural networks API (application program interface) written in Python, with TensorFlow (an open-source artificial AI library developed by Google) serving as the backend engine. The current engine-weight prediction results, together with those for the TSFC (thrust specific fuel consumption) and core-size predictions that were studied previously by the author, show that machine learning-based predictive analytics can be an effective, time-saving tool for aircraft engine design-space exploration during the conceptual design stage. It would enable expeditious identification of the best engine design amongst several candidates.

Michael T Tong↗

Cross-property deep transfer learning framework for enhanced predictive analytics on small materials data

Abstract Artificial intelligence (AI) and machine learning (ML) have been increasingly used in materials science to build predictive models and accelerate discovery. For selected properties, availability of large databases has also facilitated application of deep learning (DL) and transfer learning (TL). However, unavailability of large datasets for a majority of properties prohibits widespread application of DL/TL. We present a cross-property deep-transfer-learning framework that leverages models trained on large datasets to build models on small datasets of different properties. We test the proposed framework on 39 computational and two experimental datasets and find that the TL models with only elemental fractions as input outperform ML/DL models trained from scratch even when they are allowed to use physical attributes as input, for 27/39 (≈ 69%) computational and both the experimental datasets. We believe that the proposed framework can be widely useful to tackle the small data challenge in applying AI/ML in materials science.

36 MATERIALS SCIENCE↗

Artificial Intelligence and Machine Learning Applications in Modern Power Systems

Machine learning (ML) and artificial intelligence (AI) algorithms offer valuable tools for the analysis and interpretation of large datasets. These tools have the capability to uncover insights that may not be readily apparent within these datasets. In recent years, the integration of ML and AI has become increasingly prevalent in various applications within the power system domain. One of the earliest instances of machine learning in power systems can be traced back to demand forecasting, where artificial neural networks were employed for short-term load forecasting. In contemporary power systems, an abundance of high-resolution geospatial and temporal data is generated at various time intervals, ranging from sub-seconds (Phasor Measurement Units or PMUs) to seconds (Supervisory Control and Data Acquisition or SCADA), minutes (Process Information or PI), and extending to days, months, and years. These datasets contain valuable information concerning system reliability and performance. This information holds the potential to offer critical insights into system operations, as well as solutions for predicting and mitigating contingencies to prevent cascading outages. Despite the immense power of machine learning tools, system operators, planners, and utilities often exhibit hesitancy in fully embracing AI-enabled system operations and planning. This cautious approach persists, even as numerous diverse applications of machine learning continue to emerge in the realm of power systems. In this chapter, our focus will delve deep into ML and AI applications tailored for power systems. These applications aim to furnish system operators with enhanced situational awareness and augment their decision-making capabilities, especially during challenging operating conditions. Specific areas of interest encompass root cause analyses of electricity market datasets and the strategic selection of representative samples from vast power system databases for training ML/AI models. Finally, the chapter will conclude with a short discussion on the future of ML/AI in power systems and possible directions that the industry is moving towards.

power system applications, machine learning (ML), ↗

ChemNet: A Transferable and Generalizable Deep Neural Network for Small-Molecule Property Prediction

With access to large datasets, deep neural networks through representation learning have been able to identify patterns from raw data, achieving human-level accuracy in image and speech recognition tasks. However, in chemistry, availability of large standardized and labelled datasets is scarce, and with a multitude of chemical properties of interest, chemical data is inherently small and fragmented. In this work, we explore transfer learning techniques in conjunction with the existing Chemception CNN model, to create a transferable and generalizable deep neural network for small-molecule property prediction. Our latest model, ChemNet learns in a semi-supervised manner from inexpensive labels computed from the ChEMBL database. When fine-tuned to the Tox21, HIV and FreeSolv dataset, which are 3 separate chemical tasks that ChemNet was not originally trained on, we demonstrate that ChemNet exceeds the performance of existing Chemception models, contemporary MLP models that trains on molecular fingerprints, and it matches the performance of the ConvGraph algorithm, the current state-of-the-art. Furthermore, as ChemNet has been pre-trained on a large diverse chemical database, it can be used as a universal “plug-and-play” deep neural network, which accelerates the deployment of deep neural networks for the prediction of novel small-molecule chemical properties.

Artificial Intelligence↗

SOMAS: a platform for data-driven material discovery in redox flow battery development

Abstract Aqueous organic redox flow batteries offer an environmentally benign, tunable, and safe route to large-scale energy storage. The energy density is one of the key performance parameters of organic redox flow batteries, which critically depends on the solubility of the redox-active molecule in water. Prediction of aqueous solubility remains a challenge in chemistry. Recently, machine learning models have been developed for molecular properties prediction in chemistry and material science. The fidelity of a machine learning model critically depends on the diversity, accuracy, and abundancy of the training datasets. We build a comprehensive open access organic molecular database “Solubility of Organic Molecules in Aqueous Solution” (SOMAS) containing about 12,000 molecules that covers wider chemical and solubility regimes suitable for aqueous organic redox flow battery development efforts. In addition to experimental solubility, we also provide eight distinctive quantum descriptors including optimized geometry derived from high-throughput density functional theory calculations along with six molecular descriptors for each molecule. SOMAS builds a critical foundation for future efforts in artificial intelligence-based solubility prediction models.

25 ENERGY STORAGE↗

Advancing AI-Driven Analysis in X-ray Absorption Spectroscopy: Spectral Domain Mapping and Universal Models

In recent years, rapid progress has been made in developing artificial intelligence (AI) and machine learning (ML) methods for X-ray absorption spectroscopy (XAS) analysis. Compared to traditional XAS analysis methods, AI/ML approaches offer dramatic improvements in efficiency and help eliminate human bias. To advance this field, we advocate an AI-driven XAS analysis pipeline that features several interconnected key building blocks: benchmarks, workflows, databases, and AI/ML models. Specifically, we present two case studies for XAS ML. In the first study, we demonstrate the importance of reconciling the discrepancies between simulation and experiment using spectral domain mapping (SDM). Our ML model, which is trained solely on simulated spectra, predicts an incorrect oxidation state trend for Ti atoms in a combinatorial zinc titanate film. After transforming the experimental spectra into a simulation-like representation using SDM, the same model successfully recovers the correct oxidation state trend. In the second study, we explore the development of universal XAS ML models that are trained on the entire periodic table, which enables them to leverage common trends across elements. Looking ahead, we envision that an AI-driven pipeline can unlock the potential of real-time XAS analysis to accelerate scientific discovery.

36 MATERIALS SCIENCE↗

MPEX AI Digital Twins

All magnetically confined plasma fusion power plant concepts (Tokamak, Spherical Tokamak, Stellarator, Mirror, ...) must exhaust the heat and plasma from the core confinement region to the material walls. The primary channel for this exhaust is through a plasma divertor which directs plasma along open magnetic field lines to a material target. The Material Plasma Exposure eXperiment (MPEX) illustrated in Figure 1, is a high-power, steady-state linear plasma device designed to produce the plasma material interaction (PMI) conditions of the divertor of future magnetic confinement fusion power plants: energy flux 20MW/m 2 , ion fluence 1031/m 2 , pulse duration 106 sec. These goals of plasma exposure in MPEX are well beyond those achieved in magnetic fusion experimental devices. Successfully achieving these high power steady state conditions for long pulses requires operational control of the heating and particle sources and the plasma flux to the walls and target. The MPEX AI Hot Spot Controller, proposed in this project, will help achieve the operational milestones of MPEX. The MPEX device will begin commissioning at the end of FY26. A smaller proto-MPEX was operated for 14,666 plasma discharges and will resume operation in September of 2025 as proto-MPEX-lite, with reduced capability, to test a new window for the Helicon plasma source. The proto-MPEX data has undergone surrogate modeling with machine learning methods (R. Archibald, 2022 IEEE International Conference on Big Data). This proto-MPEX data will be used to begin development of the AI digital twins described in this white paper. The scientific mission of MPEX is to qualify materials of different composition for use in the high energy and plasma flux conditions of a fusion power plant. The materials exposed in MPEX will in some cases be exposed to high neutron fluxes at other ORNL facilities to measure the changes to their PMI properties. The targets exposed in MPEX will be transported under vacuum to a Surface Analysis Station (SAS). The SAS will be equipped with the following diagnostics: Focused Ion Beam (FIB) for trench milling, 100-400 angstrom resolution scanning electron microscope (SEM), surface mapping x-ray spectrometer, high resolution camera, and a future upgrade to a laser induced breakdown spectroscopy quadruple mass spectrometer (LIBS-QMS). The MPEX experiments will generate diverse pre- and post-exposure measurement data of detailed material properties down to the crystal grain level in 3D for post-exposure assessment of PMI damage (e.g. cracking, melting, erosion and redeposition of the material). Physics models for the PMI, and how the material composition and manufacturing impact its performance under high energy plasma exposure, need to be validated with MPEX data to guide the selection of new candidate materials. Our vision for the MPEX AI Digital Twins project is to supply experimental and physics model simulation data to train Artificial Intelligence (AI) models for data processing, analysis, operational control, PMI and materials simulation to maximize the scientific output of the MPEX device. Ultimately, an AI digital twin of MPEX material assessment metrics for tested and synthetic material types with simulated PMI will be trained by the AI Modeling Teams on the experimental and physics simulation data submitted to the American Science Cloud by this project. A purely empirical search for the best material is inefficient given the finite number of samples that can be tested on MPEX. In order to expand the material properties database for training the MPEX Material Assessment AI Digital Twin, and to gain physics understanding of the PMI processes, physics models of the material properties and PMI processes are required. The physics simulations provide detailed simulation data, like impact angles for plasma ions, sputtering yields, transport of the ionized sputtered target material in the plasma, and redeposition locations. This simulation data expands the measurement data for deeper physics understanding. The experimental data is essential to validate the PMI and material structure simulation models. The validated models can then be used to generate new simulation data of MPEX material assessments for synthetic material compositions that have not been exposed in MPEX. These predictive simulations, plus the whole experimental dataset, will be used to train the MPEX Material Assessment AI Digital Twin allowing a rapid generative AI search for new materials with reduced PMI damage by interpolating the domain of the training set. These new optimum materials can be simulated with the physics codes and/or tested in MPEX. The ability of AI neural networks to interpolate multi-dimensional parameter spaces and generate virtual data is exploited for a more efficient search for optimum materials. The advent of the Transformational AI Models Consortium (TAIMC) is an opportunity to engage with state of the art private and public AI developers to achieve the goals of the AI digital twins and AI accelerated physics models proposed in this project. Our partners at ORNL from the Advance Scientific Computing Research (ASCR) organization will collaborate in accelerating the integrated plasma material interaction simulation framework. This simulation framework will provide a platform for generating simulation data across a range of physical fidelities, including hybrid methods that produce multi-fidelity results. This data will be leveraged for AI model development, both for generation of surrogates and the automation of simulation campaigns. A part of the research below will include collaborative efforts with the TAIMC to (i) adapt data storage approaches to ensure AI-readiness, (ii) provide a protypical exemplar to inform and exercise constructed workflows, and (iii) generate and share data, using the TAIMC unified AI data standard, for foundational models that will be trained from multiple sources across the DOE complex. We will also collaborate with the TAIMC, as well as the planned AI modeling teams, to develop approaches for reducing the cost of data generation. These include tailored multi-fidelity approaches as well as fine-tuning strategies to augment general, large-scale foundational models.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Acceleration of Graph Neural Network-Based Prediction Models in Chemistry via Co-Design Optimization on Intelligence Processing Units

Atomic structure prediction and associated property calculations are the bedrock of chemical physics. Since high-fidelity ab initio modeling techniques for computing the structure and properties can be prohibitively expensive, this motivates the development of machine-learning (ML) models that make these predictions more efficiently. Training graph neural networks over large atomistic databases introduces unique computational challenges such as the need to process millions of small graphs with variable size and support communication patterns that are distinct from learning over large graphs such as social networks. We demonstrate a novel hardware-software co-design approach to scale up the training of atomistic graph neural networks (GNN) for structure and property prediction. First, to eliminate redundant computation and memory associated with alternative padding techniques and to improve throughput via minimizing communication, we formulate the effective coalescing of the batches of variable-size atomistic graphs as the bin packing problem and introduce a hardware-agnostic algorithm to pack these batches. In addition, we propose hardware-specific optimizations including a planner and vectorization for the gather-scatter operations targeted for Graphcore’s Intelligence Processing Unit (IPU), as well as model-specific optimizations such as merged communication collectives and optimized softplus. Putting these all together, we demonstrate the effectiveness of the proposed co-design approach by providing an implementation of a well-established atomistic GNN on the Graphcore IPUs. We evaluate the training performance on multiple atomistic graph databases with varying degrees of graph counts, sizes and sparsity. Here, we demonstrate that such a co-design approach can reduce the training time of atomistic GNNs and can improve the performance by up to 1.5× compared to the baseline implementation of the model on the IPUs. Additionally, we compare our IPU implementation with a Nvidia GPU-based implementation and show that our atomistic GNN implementation on the IPUs can run 1.8× faster on average compared to the execution time on the GPUs.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗