Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Neural architecture search”

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.

90 records · Page 5
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Lithium-Ion Battery Diagnostics Using Electrochemical Impedance via Machine-Learning

Diagnosing battery states such as health, state-of-charge, or temperature is crucial for ensuring the safety and reliability of electrochemical energy storage systems. While some states, such as temperature, may be measured using cheap sensors, accurate diagnosis of battery health metrics usually requires time-consuming performance measurements, making them infeasible for use in real-world operation. These health metrics can be measured during lab-testing and then estimated on-line using predictive life models or via state observer algorithms such as Kalman filters, but these predictive methods should be supplemented by actual measurement of battery health whenever possible to ensure reliability. Rapid measurement of battery health may be done by various types of fast diagnostic techniques such as electrochemical impedance spectroscopy (EIS), which can be performed in only a few minutes and require only a fraction of the energy and power needed for a full charge and discharge measurement. But there is a substantial challenge for estimating battery health using EIS data, as EIS is sensitive to cell temperature, state-of-charge, current, and resting time in addition to health. Thus, utilizing EIS data to predict battery capacity requires correcting for all these additional variables, a task that is extremely difficult to handle analytically. This talk utilizes machine-learning methods to estimate the effectiveness of battery capacity prediction from EIS data, leveraging a data set of hundreds of EIS measurements recorded at varying temperature and state-of-charge throughout a 500-day aging study of 32 commercial, large-format NMC-Graphite lithium-ion batteries. Using EIS as input to machine-learning models is complicated by the nonlinear response of impedance to battery health, temperature, and state-of-charge, as well as the collinearity between the impedance response at neighboring frequencies, which can easily lead to overfit models. To train robust models, features from EIS data need to be extracted from the data or some subset of critical frequencies selected. Many approaches for extracting and selecting features from EIS data from electrochemical analysis and machine-learning fields were identified for analysis: using the entire raw spectra; selection of one, two, or many frequencies from the entire spectra; selecting interesting points from the EIS measurement using domain knowledge; fitting EIS with an equivalent-circuit model; calculating statistics on the raw impedance values; and reducing the dimensionality of the data using unsupervised linear (principal component analysis) and non-linear (uniform manifold approximation and projection) methods. These approaches were rigorously compared using a machine-learning pipeline approach, training linear, Gaussian process, and random forest regression models and quantifying performance using cross-validation as well as a held-out test set. An artificial neural network model trained on the raw spectra was also tested. Promising pipelines were fine-tuned via Bayesian hyperparameter optimization using cross-validation loss and training with class-specific weights to counter data set imbalance. The most reliable method for utilizing impedance in this work was the selection of two optimal frequencies through an exhaustive search, resulting in about 2% mean absolute error on test data for both Gaussian process and random forest model architectures. Interrogation of a variety of models reveals critical frequencies of 100 Hz and 103 Hz for this data set, though the optimal set of frequencies is not necessarily intuitive, i.e., the best performing models are not simply those that use impedance at frequencies that have the highest correlation to the relative discharge capacity. The best performing model is an ensemble model, which is able to predict battery capacity with 1.9% mean absolute error for unseen cells using impedance recorded at a variety of temperatures and states-of-charge.

battery↗
DOE OSTI2024

Equivariant, safe and sensitive — graph networks for new physics

This study introduces a novel Graph Neural Network (GNN) architecture that leverages infrared and collinear (IRC) safety and equivariance to enhance the analysis of collider data for Beyond the Standard Model (BSM) discoveries. By integrating equivariance in the rapidity-azimuth plane with IRC-safe principles, our model significantly reduces computational overhead while ensuring theoretical consistency in identifying BSM scenarios amidst Quantum Chromodynamics backgrounds. The proposed GNN architecture demonstrates superior performance in tagging semi-visible jets, highlighting its potential as a robust tool for advancing BSM search strategies at high-energy colliders.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2022

Prediction of DIII-D Pedestal Structure from Externally Controllable Parameters

The sharp increase of pressure at the edge of a high confinement mode (H-mode) plasma, the pedestal, strongly impacts overall plasma performance. Predicting the pedestal is a necessity to control and optimize tokamak operations. An experimental data-driven machine learning (ML) approach is presented that predicts the pedestal heights and widths of electron density (ne) and electron temperature (Te) profiles as well as the separatrix ne from externally controllable parameters such as the plasma shape, heating method and power, and gas puff rate and integrated gas puff. The OMFIT framework was used with DIII-D data to efficiently, robustly, and automatically build a database of pedestal parameters to train machine learning models. Database creation was enabled by the search engine tool for DIII-D data, TokSearch, which parallelizes data fetching, enabling fast searches through basic signals of thousands of DIII-D shots and selection of relevant time intervals. Principal Component Analysis (PCA) separated the database into three clusters that represent classes of plasma shapes that are regularly used in DIII-D. The most important parameters for setting the pedestal structure were plasma current (Ip), toroidal magnetic field (Bφ), neutral beam heating power (PNBI) and shaping quantities. The Deep Jointly Informed Neural Networks (DJINN) algorithm was applied to identify suitable neural network (NN) architectures that appropriately capture the features of the pedestal database. Separate NNs were implemented for each pedestal parameter, and ensembling methods were used to improve the prediction accuracy and allowed estimation of the prediction uncertainty. The pedestal predictions of the test dataset lie within the measurement uncertainties of the pedestal parameters. The NN outperformed simple Linear Regression (LR) analysis, indicating non-linear dependencies in the pedestal structure. The presented achievements illustrate a promising path for future research, using feature extraction to infer experimental trends and thereby improve pedestal models as well as deploying NN for a fast pedestal prediction in DIII-D scenario development.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2021

Prediction of DIII-D Pedestal Structure From Externally Controllable Parameters

The sharp increase of pressure at the edge of a high confinement mode (H-mode) plasma, the pedestal, strongly impacts overall plasma performance. Predicting the pedestal is a necessity to control and optimize tokamak operations. Here, an experimental data-driven machine learning (ML) approach is presented that predicts the pedestal heights and widths of electron density (n e ) and electron temperature (T e ) profiles as well as the separatrix ne from externally controllable parameters such as the plasma shape, heating method and power, and gas puff rate and integrated gas puff. The OMFIT framework was used with DIII-D data to efficiently, robustly, and automatically build a database of pedestal parameters to train machine learning models. Database creation was enabled by the search engine tool for DIII-D data, TokSearch, which parallelizes data fetching, enabling fast searches through basic signals of thousands of DIII-D shots and selection of relevant time intervals. Principal Component Analysis (PCA) separated the database into three clusters that represent classes of plasma shapes that are regularly used in DIII-D. The most important parameters for setting the pedestal structure were plasma current (I p ), toroidal magnetic field (B Φ ), neutral beam heating power (P NBI ) and shaping quantities. The Deep Jointly Informed Neural Networks (DJINN) algorithm was applied to identify suitable neural network (NN) architectures that appropriately capture the features of the pedestal database. Separate NNs were implemented for each pedestal parameter, and ensembling methods were used to improve the prediction accuracy and allowed estimation of the prediction uncertainty. The pedestal predictions of the test dataset lie within the measurement uncertainties of the pedestal parameters. The NN outperformed simple Linear Regression (LR) analysis, indicating non-linear dependencies in the pedestal structure. The presented achievements illustrate a promising path for future research, using feature extraction to infer experimental trends and thereby improve pedestal models as well as deploying NN for a fast pedestal prediction in DIII-D scenario development.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2026

Discovering Strong Gravitational Lenses in the Dark Energy Survey with Interactive Machine Learning and Crowd-sourced Inspection with Space Warps

We conduct a search for strong gravitational lenses in the Dark Energy Survey (DES) Year 6 imaging data. We implement a pre-trained Vision Transformer (ViT) for our machine learning (ML) architecture and adopt interactive machine learning to construct a training sample with multiple classes to address common types of false positives. Our ML model reduces ∼236 million DES cutout images to 22,564 targets of interest, including ∼85% of previously reported galaxy–galaxy lens candidates discovered in DES. These targets were visually inspected by citizen scientists, who ruled out ∼90% as false positives. Of the remaining 2618 candidates, 149 were expert-classified as “definite” lenses and 516 as “probable” lenses, for a total of 665 systems, with 147 of these candidates being newly identified. Additionally, we trained a second ViT to find double-source plane lens systems, finding at least one double-source system. Our main ViT excels at identifying galaxy–galaxy lenses, consistently assigning high scores to candidates with high expert assessments. The top 800 ViT-scored images include ∼100 of our “definite” lens candidates. This selection is an order of magnitude higher in purity than previous convolutional neural-network-based lens searches and demonstrates the feasibility of applying our methodology for discovering large samples of lenses in future surveys.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2022

Automatic Generation of High-Performance Convolution Kernels on ARM CPUs for Deep Learning

In this work, we present FastConv, a template-based code auto-generation open source library that can automatically generate high-performance deep learning convolution kernels of arbitrary matrices/tensors shapes. FastConv is based on the Winograd algorithm, which is reportedly the highest performing algorithm for the time-consuming convolution layers of convolutional neural networks. ARM CPUs cover a wide range designs and specifications, from embedded devices to HPC-grade CPUs. The leads to the dilemma of how to consistently optimize Winograd-based convolution solvers for convolution layers of different shapes. FastConv addresses this problem by using templates to auto-generate multiple shapes of tuned kernels variants suitable for skinny tall matrices. As a performance portable library, FastConv transparently searches for the best combination of kernel shapes, cache tiles, scheduling of loop orders, packing strategies, access patterns, and online/offline computations. Auto-tuning is used to search the parameter configuration space for the best performance for a given target architecture and problem size. The experiments with layer-wise evaluation on the VGG--16 model confirms a 1.25x performance gains is got by tuning the Winograd library. Integrated comparison results shows 1.02x to 1.40x, 1.14x to 2.17x, and 1.22x and 2.48x speedup is achieved over NNPACK, Arm NN, and FeatherCNN on the Kunpeng 920 beside few cases. Furthermore, problem size performance portability experiments with various convolution shapes shows that FastConv achieves 1.2x to 1.7x speedup and 2x to 22x speedup over NNPACK and ARM NN inference engine using Winograd on Kunpeng 920 . CPU performance portability evaluation on the VGG--16 show an average speedup over NNPACK of 1.42x, 1.21x, 1.26x, 1.37x, 2.26x, and 11.02x is observed on Kunpeng 920, Snapdragon 835, 855, 888, Apple M1, and AWS Graviton2, respectively.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Optimizing the optimizer for physics-informed neural networks and Kolmogorov-Arnold networks

Physics-Informed Neural Networks (PINNs) have revolutionized the computation of PDE solutions by integrating partial differential equations (PDEs) into the neural network’s training process as soft constraints, becoming an important component of the scientific machine learning (SciML) ecosystem. More recently, physics-informed Kolmogorv-Arnold networks (PIKANs) have also shown to be effective and comparable in accuracy with PINNs. In their current implementation, both PINNs and PIKANs are mainly optimized using first-order methods like Adam, as well as quasi-Newton methods such as BFGS and its low-memory variant, L-BFGS. However, these optimizers often struggle with highly nonlinear and non-convex loss landscapes, leading to challenges such as slow convergence, local minima entrapment, and (non)degenerate saddle points. In this study, we investigate the performance of Self- Scaled BFGS (SSBFGS), Self-Scaled Broyden (SSBroyden) methods and other advanced quasi-Newton schemes, including BFGS and L-BFGS with different line search strategies. These methods dynamically rescale updates based on historical gradient information, thus enhancing training efficiency and accuracy. We systematically compare these optimizers – using both PINNs and PIKANs – on key challenging PDEs, including the Burgers, Allen-Cahn, Kuramoto-Sivashinsky, Ginzburg-Landau, and Stokes equations. Additionally, we evaluate the performance of SSBFGS and SSBroyden for Deep Operator Network (DeepONet) architectures, demonstrating their effectiveness for data-driven operator learning. Our findings provide state-of-the-art results with orders-of-magnitude accuracy improvements without the use of adaptive weights or any other enhancements typically employed in PINNs. More broadly, our work reveal insights into the effectiveness of quasi-Newton optimization strategies in significantly improving the convergence and accurate generalization of PINNs and PIKANs.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Bayesian Optimized Deep Ensemble for Uncertainty Quantification of Deep Neural Networks: a System Safety Case Study on Sodium Fast Reactor Thermal Stratification Modeling

Deep neural networks (DNNs) are increasingly important to scientific computing and engineering system simulations. Accurate uncertainty quantification (UQ) for DNNs is critical in safety-sensitive engineering domains. Traditional Deep Ensemble (DE) methods, while easy to implement, frequently suffer from poorly calibrated uncertainty estimates and limited predictive accuracy due to reliance on fixed architectures with varied weight initializations. To address these issues, we introduce a workflow that combines Bayesian Optimization (BO) and DE. The workflow is modular, scalable, and integrates parallel BO initialized with Sobol sequences to individually optimize the hyperparameters of each ensemble member. This method enhances ensemble diversity, improves predictive accuracy, and provides reliable uncertainty estimates. We evaluate the proposed BODE approach in a sodium fast reactor thermal stratification modeling case study, where we used a densely connected convolutional neural network to predict turbulent viscosity during the reactor transient with consideration of data noise. We benchmark its performance against several optimization approaches, including baseline deep ensemble, evolutionary algorithm-optimized ensemble, ensemble formed via random search combined with greedy selection, and a BO ensemble using random initialization. Here, our results demonstrate superior performance of the developed BODE approach. In noise-free scenarios, BODE notably reduces incorrect aleatoric uncertainty and significantly enhances predictive accuracy. Under conditions of 5% and 10% Gaussian noise, BODE adaptively quantifies uncertainty proportional to data noise, achieving up to an 80% reduction in root mean square error compared to baseline methods and producing well-calibrated prediction intervals.

Bayesian optimization↗
DOE OSTI2021

Stopping criteria for ending autonomous, single detector radiological source searches

While the localization of radiological sources has traditionally been handled with statistical algorithms, such a task can be augmented with advanced machine learning methodologies. The combination of deep and reinforcement learning has provided learning-based navigation to autonomous, single-detector, mobile systems. However, these approaches lacked the capacity to terminate a surveying/search task without outside influence of an operator or perfect knowledge of source location (defeating the purpose of such a system). Two stopping criteria are investigated in this work for a machine learning navigated system: one based upon Bayesian and maximum likelihood estimation (MLE) strategies commonly used in source localization, and a second providing the navigational machine learning network with a “stop search” action. A convolutional neural network was trained via reinforcement learning in a 10 m × 10 m simulated environment to navigate a randomly placed detector-agent to a randomly placed source of varied strength (stopping with perfect knowledge during training). The network agent could move in one of four directions (up, down, left, right) after taking a 1 s count measurement at the current location. During testing, the stopping criteria for this navigational algorithm was based upon a Bayesian likelihood estimation technique of source presence, updating this likelihood after each step, and terminating once the confidence of the source being in a single location exceeded 0.9. A second network was trained and tested with similar architecture as the previous but which contained a fifth action: for self-stopping. The accuracy and speed of localization with set detector and source initializations were compared over 50 trials of MLE-Bayesian approach and 1000 trials of the CNN with self-stopping. The statistical stopping condition yielded a median localization error of ~1.41 m and median localization speed of 12 steps. The machine learning stopping condition yielded a median localization error of 0 m and median localization speed of 17 steps. This work demonstrated two stopping criteria available to a machine learning guided, source localization system.

38 RADIATION CHEMISTRY, RADIOCHEMISTRY, AND NUCLEA↗
NASA NTRS1990

Programmable synaptic devices for electronic neural nets

The architecture, design, and operational characteristics of custom VLSI and thin film synaptic devices are described. The devices include CMOS-based synaptic chips containing 1024 reprogrammable synapses with a 6-bit dynamic range, and nonvolatile, write-once, binary synaptic arrays based on memory switching in hydrogenated amorphous silicon films. Their suitability for embodiment of fully parallel and analog neural hardware is discussed. Specifically, a neural network solution to an assignment problem of combinatorial global optimization, implemented in fully parallel hardware using the synaptic chips, is described. The network's ability to provide optimal and near optimal solutions over a time scale of few neuron time constants has been demonstrated and suggests a speedup improvement of several orders of magnitude over conventional search methods.

Moopenn, A.↗
NASA NTRSDate not supplied

Deep Learning-Based Negotiation Strategy Selection for Cooperative Conflict Resolution in Urban Air Mobility

This paper presents a collaborative conflict resolution technique using deep neural network-based intelligent search of the solution space. This approach offers a rapid convergence to a mutually acceptable solution for real-time conflict resolution, suitable for urban air mobility operations. Furthermore, the presented technique allows operational flexibility to the urban air mobility agents where these agents can collaboratively devise the solution via integrative negotiation, based on their local utility functions, as long as such a solution does not violate the global safety thresholds. The presented machine-to-machine negotiation method is built on our prior work on holistic assessment of the airspace and potential conflict detection implemented at-the-edge, onboard the unmanned aircraft systems. This paper extends the prior work to augment decision-making at-the-edge, thereby, promising a true distributed control architecture for urban air mobility. In this approach, each agent (a) builds a potential in-flight conflict map, (b) identifies the conflicting agents, (c) dynamically prepares a list of alternatives based on its current utility functions, (d) negotiates with the conflicting agents to pick one of these alternatives, and (e) implements the negotiated alternative to mutually resolve the conflict. Note that such an approach does not require a contingency plan to be made pre-flight, as the conflict resolution strategies are decided and negotiated in real time based on the present state of the agent. The contingency plan, if available, can serve as an input to the real-time conflict resolution strategy formulation, and also can be used as a fallback plan in case the negotiation fails and the impacted agents need to switch to a rule-based/supervisory resolution mode from the discussed distributed resolution mode. The presented collaborative negotiation-based conflict resolution technique incorporates a time-dependent reward function to catalyze collaborative resolution by incentivizing the agents with local and global rewards beneficial to their business operations.

Advanced Air Mobility↗
DOE OSTI2025

Robustness of Deep Learning Classification to Adversarial Input on GPUs: Asynchronous Parallel Accumulation Is a Source of Vulnerability

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.

Shanmugavelu, Sanjif [Maxeler Technologies, a Groq↗
DOE OSTI2022

Exploring the Use of Novel Spatial Accelerators in Scientific Applications

Driven by the need to find alternative accelerators which can viably replace GPUs in next-generation Supercomputing systems, this paper proposes a methodology to enable agile application/hardware co-design. The application-first methodology provides the ability to come up with design of accelerators while working with real-world workloads, available accelerators, and system software. The iterative design process targets a set of kernels in a workload for performance estimates that can prune the design space for later phases of detailed architectural evaluations. To this effect, in this paper, a novel data-parallel device model is introduced that simulates the latency of performance-sensitive operations in an accelerator including data transfers and kernel computation using multi-core CPUs. The use of off-the-shelf simulators, such as pre-RTL simulator Aladdin or multiple tools available for exploring the design of deep neural network accelerators (e.g., Timeloop) is demonstrated for evaluation of various accelerator designs using applications with realistic inputs. Examples of multiple device configurations that are instantiable in a system are explored to evaluate the performance benefit of deploying novel accelerators. The proposed device is integrated with a programming model and system software to potentially explore the impacts of high-level programming languages/compilers and low-level effects such as task scheduling on multiple accelerators. We analyze our methodology for a set of applications that represent high-performance computing (HPC) and graph analytics. The applications include a computational chemistry kernel realized using tensor contractions, triangle counting, GraphSAGE and Breadth-first Search. These applications include kernels such as dense matrix-dense matrix multiplication, sparse matrix-spare matrix multiplication, and sparse matrix-dense vector multiplication. Our results indicate potential performance benefits and insights for system design by including accelerators that realize these kernels along-side general purpose accelerators.

AI, codesign, Accelerated Computing, Modeling and ↗
NASA NTRS1991

Electronic neural network for dynamic resource allocation

A VLSI implementable neural network architecture for dynamic assignment is presented. The resource allocation problems involve assigning members of one set (e.g. resources) to those of another (e.g. consumers) such that the global 'cost' of the associations is minimized. The network consists of a matrix of sigmoidal processing elements (neurons), where the rows of the matrix represent resources and columns represent consumers. Unlike previous neural implementations, however, association costs are applied directly to the neurons, reducing connectivity of the network to VLSI-compatible 0 (number of neurons). Each row (and column) has an additional neuron associated with it to independently oversee activations of all the neurons in each row (and each column), providing a programmable 'k-winner-take-all' function. This function simultaneously enforces blocking (excitatory/inhibitory) constraints during convergence to control the number of active elements in each row and column within desired boundary conditions. Simulations show that the network, when implemented in fully parallel VLSI hardware, offers optimal (or near-optimal) solutions within only a fraction of a millisecond, for problems up to 128 resources and 128 consumers, orders of magnitude faster than conventional computing or heuristic search methods.

Thakoor, A. P.↗
NASA NTRSDate not supplied

Development of Machine Learning Algorithms to Segment and Study Images of Astromaterial Samples

Introduction: Micrometer-scale chemical analyses of chondritic meteorites and mission-returned asteroid samples can reveal details of the physical and chemical processes operating in the early solar system, including processes that gave rise to planets, moons, and minor bodies. These primitive astromaterials are comprised of chondrules, calcium- and aluminum-rich inclusions (CAI), and many other silicates, oxides, metals, sulfides, and fine-grained materials. The chemical and mineralogical complexity of these samples, vast populations of different components, and heterogeneity across mm to km scales, all limit our understanding of the origin and evolution of these materials. Here, we describe recent efforts to use machine learning techniques to automate the segmentation of chemical maps of chondritic meteorites, designed to aid studies of asteroid samples returned by spacecraft. By automating the task of segmentation it will become possible to rapidly analyze and interpret the sizes, shapes, mineralogy, chemistry, and other properties of every chondrule, calcium- and aluminum-rich inclusion (CAI) and other clast within and between asteroid samples. Sample return missions significantly accelerate and heighten the need to develop such new data analysis techniques, and associated data repositories. Techniques: Neural networks require abundant training data, i.e. images which have been segmented by a human user. We have manually segmented data available from previous petrologic and chemical work at NASA Johnson Space Center and the American Museum of Natural History [1-4]. These data were derived from energy- and wavelength-dispersive X-ray spectroscopy (EDS, WDS) mapping of samples from many chondrite groups. The Deeplabv3+ [5] neural network architecture was trained on human-labeled masks and used to create machine-labeled masks. Several different algorithms were investigated, with inputs ranging from common RGB image formats through to hyperspectral datasets, with raw data comprising greyscale maps of Mg, Ca, and Al, with or without Si, Fe, Ti for both EDS and WDS data, and extending to other elements in EDS only. Each greyscale image was paired with a binary mask for each labelled particle type. Results: The trained algorithms can segment (Fig 1), classify, and measure the dimensions of thousands of particles in chemical maps of a standard 1-inch round petrographic section in seconds to minutes, rather than many hours needed by a human. Accuracy of the algorithms varied from chondrite to chondrite and across particle types. Further results and details of the algorithms will be presented at the workshop. Future directions: Machine learning has the potential to revolutionize our understanding of complex particle populations contained within primitive astromaterial, with segmentation being a critical first step. Example applications include better understanding of particle transport, nebular reservoirs, parent body accretion, and a deeper understanding of the relationships between particle populations and bulk rock elemental and isotopic compositions. In addition to benefits that machine learning can bring to individual researchers, building a community data repository of thousands to millions of particles across hundreds of samples will open up many other possibilities. For example, with a large enough dataset it will be possible to search for exceptionally closely matching particles across disparate samples. Such a capability would enable a single CAI from OSIRISREx or Hayabusa/II samples to be matched to chondritic CAIs that exhibit near-identical size, texture, and mineralogy, down to the level of similar core phenocrysts, zonation, and rim sequences. Such comparative analyses will help to disentangle precursor chemistry, chronology, gas/dust reservoirs during heating, and accretion. Such an endeavor would be impossible without machine learning and a large community data repository of astromaterial chemical/mineralogic maps.

Machine Learning↗
NASA NTRSDate not supplied

Enhancing Air Traffic Control Planning with Automatic Speech Recognition

The decisions made during the Federal Aviation Administration Air Traffic Control System Command Center's planning teleconferences hold significant sway over the National Airspace System. Held every two hours, these teleconferences convene air traffic managers and stakeholders from across the nation to discuss airspace conditions, weather, and constraints, leading to the formulation and adjustment of traffic management initiatives. Given the critical nature of these decisions, the need for accurate and efficient record-keeping is paramount. In recent years, the application of automatic speech recognition has gained popularity across diverse industries, including aviation. While traditional applications focus on transcribing air traffic control communication, this paper explores a unique application of automatic speech recognition by converting the audio from planning teleconferences into text transcriptions. This innovative approach addresses key challenges in the field, presenting potential benefits for quality assurance, real-time participation, and downstream natural language processing tasks. A notable breakthrough in the machine learning community, namely the transformer neural network architecture, forms the backbone of the proposed solution in this paper. The transformer architecture's role in this research represents a paradigm shift in the efficiency of automatic speech recognition models. By reducing the amount of in-domain training data required, this architecture allows for the fine-tuning of such models like Whisper, originally pretrained on vast English speech datasets. The adaptability of the transformer architecture proves invaluable in capturing the nuances of aviation terminology and specific language used in planning teleconferences. Leveraging the Whisper model as a baseline, our research details the fine-tuning and validation using a dataset comprising 20 hours of meticulously transcribed planning teleconferences. Notably, the baseline pretrained Whisper model exhibited a word error rate of 18.77%. Through the fine-tuning process, the model achieved a substantial improvement, demonstrating an impressive performance with a reduced word error rate of 6.82%. This substantial decrease in WER not only highlights the effectiveness of the transformer architecture but also emphasizes the practical advancements achieved through the application of automatic speech recognition in this specific domain. The utilization of automatic speech recognition in planning teleconferences in this work introduces several novelties. Firstly, the creation of text transcriptions offers a valuable tool for quality assurance and facilitates the efficient review of teleconferences. This is an important aspect of the proposed solution, given the time-sensitive and high-stakes nature of decisions made during these meetings. Furthermore, text-searchable transcriptions provide a streamlined approach for locating and validating critical information, potentially saving hours of manual effort in searching through audio recordings. Moreover, our research identifies a key use case for external facilities and stakeholders. In situations where attendance at the planning teleconference is not feasible, having access to text transcriptions in real-time or shortly after the teleconference ends, proves to be a time-saving and informative resource. This feature enhances collaboration and ensures that stakeholders can stay abreast of important discussions and decisions even in their absence. Despite the efficiency gains facilitated by the transformer architecture in automatic speech recognition technology, it is essential to acknowledge the human factors in data creation. Subject matter experts play a crucial role in accurately transcribing planning teleconferences due to the specificity and complexity of the information discussed. The research dataset, consisting of 20 hours of transcribed planning teleconferences, forms the foundation for fine-tuning and validating the Whisper model. The achieved word error rate of 6.82% demonstrates promising advancements, particularly in recognizing essential aviation terminology within the teleconferences. In conclusion, this paper presents a comprehensive exploration of the application of automatic speech recognition in Air Traffic Control System Command Center planning teleconferences, leveraging the transformer architecture for enhanced efficiency. The novel contributions lie in the improved accessibility of decision-making records, real-time participation opportunities for external stakeholders, and the potential for downstream natural language processing advancements. As the aviation industry continues to evolve, the integration of automatic speech recognition technologies holds the promise of revolutionizing decision-making processes and contributing to the overall safety and efficiency of air traffic management.

ATM↗
DOE OSTI2020

Neural MUSE Analysis

Researchers at Oak Ridge National Laboratory (ORNL) created data as part of the MUSE (Multi-Agency Urban Search Experiment Detector and Algorithm Test Bed) project simulating illicit nuclear materials located in various buildings along a road. In the simulation, a truck containing a radiation detector drives down the road gathering listmode data (counting the and energy of incident gamma radiation). Building materials, source shielding, driving speed, truck direction, truck location on the road, source type, and source placement are all varied between runs of the data set. This data was created using deterministic neutron transport and Monte Carlo methods through a combination of SCALE, MAVRIC, MCNP, and GADRAS. As part of a follow-on NA-22 project, two Kaggle competitions were created to determine the best algorithms for finding and identifying gamma sources in this simulated urban environment. The winning algorithm was neural network-based and had a test accuracy of 76.4% accuracy for source identification. This work seeks to build upon this work and improve the results through the application of novel machine learning techniques. As a first step, the data was classified by a simple Convolutional Neural Network (CNN) To accomplish this, the data was first preprocessed into “waterfall plots.” These plots are composed of energy vs count plots that are stacked vertically to show progression in time. The horizontal axis indicating the particle energy incorporated user defined bin spacing with options for in linear-, logarithmic-, square root-, and user-spaced bins. The z or color dimension showed the number of counts corresponding the energy-time combination. This data was then used to generate more data, by generating a local estimate of the mean of the distribution for a bin and then randomly re-sampling that bin from a Poisson distribution. Once all of this data was generated, it was fed into a well-known CNN architecture, ResNet50. The output layer of this model was removed and replaced with layers corresponding to the shape desired isotope outputs. The provided training data was used to train the classifier and the remaining testing data was used to evaluate the model. Results are soon to be forthcoming.

61 RADIATION PROTECTION AND DOSIMETRY↗
NASA NTRSDate not supplied

Quantum-Assisted Variational Segmentation for Image-to-Image Wildfire Detection Using Satellite Data

The quantum computing community has been searching for suitable applications to demonstrate the potential of near-term quantum devices. Quantum machine learning is a potential candidate, particularly using models that cannot be efficiently simulated with classical computers [1, 2]. This work focuses on a transition phase of quantum computers where the quantum machine learning model is still simulable classically but projected not to be simulable as the size of the model grows. Ultimately quantum computers may have advantages for high-dimensional real-world problems. Due to the limited number of qubits in current noisy intermediate-scale quantum (NISQ) devices, the direct application of quantum computers in high dimensional data is not feasible. To remedy this problem, an encoder-decoder architecture can be utilized. The encoder model would transform the high-dimensional data into a compact representation, to a level that small quantum computers can be used today (or in the near future), and the decoder would take the quantum processed outputs back to the high-dimensional space. Addressing the two challenges of quantum machine learning, this work investigates a hybrid supervised generative model with a quantum Ising Born machine embedded as the latent distribution. The model contains four main parts (Figure 1.a.): (1) a U-NET architecture responsible for learning segmentation flow, (2) a Prior network responsible for learning an encoded latent distribution of the input data, (3) a Born machine which represents the latent distribution, and (4) a Posterior network in charge of learning the joint encoded latent distribution of inputs and target data. The initial model, proposed by [3], is optimized by (1) maximizing the overlap of the prior and posterior latent distributions, and (2) minimizing the segmentation loss. The proposed model is designed to be investigated in a simulation environment applied to the real-world application of wildfire segmentation. Specifically, the model is designed to solve the patchy wildfire segmentations of Moderate Resolution Imaging Spectroradiometer (MODIS) by taking the MODIS observations and using Visible Infrared Imaging Radiometer Suite’s (VIIRS) consistent wildfire product as the target. The model solves patchy wildfire segmentations and provides insight into the epistemic errors sourced from model variation. The model utilizes the Born machine as a QUBO solver to represent the latent space as a Bernoulli distribution. The proposed configuration allows the variational segmentation model to leverage the true quantum probabilistic nature and derive a more expressive latent configuration, increasing the model performance in describing wildfire segmentations. The quantum probabilistic information of the Born machine is directly incorporated in the Kullback-Leibler divergence loss in the prior and posterior distributions, forcing the Bernoulli latent distribution to maximize the overlap of input and joint input-target distributions. The proposed model is then trained and compared with a baseline only consisting of direct Bernoulli latent distribution with no Born machine representing the latent space. The models are evaluated based on the segmentation metrics, such as precision, recall, intersect of union, with uncertainty boundaries accounting for the stochastic nature of the model. Our findings show that even in low latent-dimensional space (due to the limit in computational power of the classical quantum simulator), we are able to effectively capture the latent representation and hence the model performs better than the baseline. The findings are a projection for scaling the model into higher dimensional latent space with the Born machine surpassing the baseline performance. Figure 1. Sub-figure (a) demonstrates the architecture for the training phase. The model consists of a Prior and Posterior network that encode inputs and joint input-target data into compact representations, respectively. The Born machine represents the latent distribution, and the U-NET branch learns the segmentation patterns of the data. The stochasticity is introduced to the U-NET through its last layer to create meaningful but stochastic segmentations. Sub-figure (b) represents the inference phase where the model takes the stochastic behavior from the prior network and injects that into the U-NET. Each attempt of inference will generate different but similar segmentations from the same distribution of the wildfire event. REFERENCES [1] Coyle, B., Mills, D., Danos, V., & Kashefi, E. (2020). The Born supremacy: quantum advantage and training of an Ising Born machine. npj Quantum Information, 6(1), 1-11. [2] Liu, J. G., & Wang, L. (2018). Differentiable learning of quantum circuit born machines. Physical Review A, 98(6), 062324. [3] Kohl, S., Romera-Paredes, B., Meyer, C., De Fauw, J., Ledsam, J. R., Maier-Hein, K., ... & Ronneberger, O. (2018). A probabilistic u-net for segmentation of ambiguous images. Advances in neural information processing systems, 31.

quantum machine learning↗