Search NASA⌕ Search

SEARCH · Search NASA

Results for “Range Of Learning”

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 73 records · Page 4
All sourcesNASA NTRSDOE OSTI
DOE OSTI2019

Multi-Formalism Modeling for Disaster Resilience, Forecasting, and Response

Disaster response preparedness, risk mitigation, and effective post-event response are critical for national security, public health, and economic integrity. For areas impacted by extreme events, timely and accurate impact assessments of lifeline infrastructures are critical. Traditional disaster management and situational awareness platforms are not capable of handling the massive influx of available information from traditional and non-traditional sources. Next-generation analytic assessment approaches, where use of dynamically sourced, heterogeneous data with near real-time availability are integrated into a continually-updated analytics framework. The objective of this research was to develop a foundational multi-formalism modeling platform to integrate novel data sources for use in improving situational awareness after a disaster event. Specifically, to 1) provide an architecture for machine-learning based data fusion various multi-sensor datasets; 2) classify, geolocate, and temporally bin novel data sources for use in spatially-enabled machine-learning based data fusion; and 3) derive confidence-based observation data with machine-learning based prediction to fill spatial and temporal observation gaps. This project developed a reusable cloud-based pipeline architecture for ingesting, storing, and analyzing relevant data, executing a range of deep-learning models, and visualize the information geographically. The pipeline development was informed by previous PNNL investments (StreamKit) and built upon leading-edge technologies consisting of: 1) a raw datastore; 2) a data enrichment pipeline with custom and third-party analytics; 3) enriched data storage; and 4) a custom UI to display the enriched data. Machine learning/deep learning models were selected and tested to classify social media images for relevancy and presence of flood waters. Our custom trained relevance models were run on a validation dataset and the overall accuracy of the relevance classifiers ranged from 65% to 90%. We found that convolutional neural network (CNN) models were both faster and more accurate than more traditional machine learning models. We applied the relevance classifier to remove non-relevant images and then undertook a limited training of CNN models to detect flood on ~1,800 labeled images. The testing accuracy of the flood detection models were ~88%. To improve the geographic location of both social media and oblique aerial images we tested a multi-step workflow built on a pre-trained deep-learning model. To identify similarity features from test and reference scenes, we applied the Xception and Faiss models to a set of test images with inaccurate spatial locations, as well and precisely-located reference scenes from the general area. Matching reference scenes to the test scenes with the closest similarity index had limited success in improving the precision of test scene location and additional methods have been identified for testing. Data fusion between authoritative geographic data (satellite imaging) and non-authoritative crowdsourced data is an underlying theme throughout this research and methods are tested bring these data to a common geographic form through feature matching and transformations. The automated extraction of supporting information for disaster response from oblique aerial images or ground-level social media images is accomplished through semantic image segmentation and transformation. The non-authoritative data are used to validate the authoritative data collections and processing and optionally, t+0 flood models. In addition, the common form geographic data are combined with other static landscape features to develop spatially-explicit training vectors used in deep generative statistical inference models to develop complex non-linear relationships in static and dynamic data and ultimately derive a spatially continuous and probabilistic assessment of flood damage with the event domain.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗
DOE OSTI2021

Predicting measures of soil health using the microbiome and supervised machine learning

Soil health encompasses a range of biological, chemical, and physical soil properties that sustain the commercial and ecological value of agroecosystems. Monitoring soil health requires a comprehensive set of diagnostics that can be cost-prohibitive for routine analyses. The soil microbiome provides a rich source of information about soil properties, which can be assayed in a high-throughput, cost-effective way. We evaluated the accuracy of random forest (RF) and support vector machine (SVM) regression and classification models in predicting 12 measures of soil health, tillage status, and soil texture from 16S rRNA gene amplicon data with an operationally relevant sample set. We validated the efficacy of the best performing models against independent datasets and also tested best practices for processing microbiome data for use in machine learning. Soil health metrics could be predicted from microbiome data with the best models achieving a Kappa value of ~0.65, for categorical assessments, and a R2 value of ~0.8, for numerical scores. Biological health ratings were better predicted than chemical or physical ratings. Validation with independent datasets revealed that models had general predictive value for soil properties, including yield. The ecological profiles of several taxa important for model accuracy matched the observed relationships with soil health, including Pyrinomonadaceae, Nitrososphaeraceae, and Candidatus Udeaobacter. Models trained at the highest taxonomic resolution proved most accurate, with losses in accuracy resulting from rarefying, sparsity filtering, and aggregating at higher taxonomic ranks. Furthermore, our study provides the groundwork for developing scalable technology to use microbiome-based diagnostics for the assessment of soil health.

16S rRNA gene↗
DOE OSTI2018

Empirical modeling of dopability in diamond-like semiconductors

Carrier concentration optimization has been an enduring challenge when developing newly discovered semiconductors for applications (e.g., thermoelectrics, transparent conductors, photovoltaics). This barrier has been particularly pernicious in the realm of high-throughput property prediction, where the carrier concentration is often assumed to be a free parameter and the limits are not predicted due to the high computational cost. In this work, we explore the application of machine learning for high-throughput carrier concentration range prediction. Bounding the model within diamond-like semiconductors, the learning set was developed from experimental carrier concentration data on 127 compounds ranging from unary to quaternary. The data were analyzed using various statistical and machine learning methods. Accurate predictions of carrier concentration ranges in diamond-like semiconductors are made within approximately one order of magnitude on average across both p- and n-type dopability. The model fit to empirical data is analyzed to understand what drives trends in carrier concentration and compared with previous computational efforts. Lastly, dopability predictions from this model are combined with high-throughput quality factor predictions to identify promising thermoelectric materials.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2025

Enhancing Short-Range Weather Forecasts through Temporal Variation Encoding: A Multiperiod Embedding Approach

Machine learning (ML) techniques have emerged as promising approaches to improve regional weather forecast accuracy and reliability through data-driven methods. We propose a novel ML-based weather forecasting model, the Multiperiod Embed Net (MPENet). A key distinguishing feature of MPENet is its explicit utilization of the inherent cyclic nature in weather dynamics, unlike the autoregressive strategies commonly used in other ML weather forecasting approaches. Critical cyclic structures are identified via Fourier analyses of dynamic time series. Cyclicity in the convolutional representation is achieved by transforming one-dimensional time series of meteorological variables into two-dimensional tensors based on identified periods. This approach enables the model to leverage intrinsic weather patterns, enhancing regional forecast performance. To demonstrate the effectiveness of MPENet, we conduct a comparative analysis with Nvidia’s FourCastNet. Both models are trained on High-Resolution Rapid Refresh (HRRR) data from 2015 to 2022, over a 192 km × 192 km region in Tennessee. The comparisons are performed locally at two specific locations known to have different weather dynamics due to orographic effects: Crossville, on the relatively flat Cumberland Plateau with fewer topographic airflow disruptions, and Oak Ridge, in the ridge-and-valley region, where airflow is heavily influenced by surrounding valleys and mountains. Our results indicate that FourCastNet achieves strong accuracy at very short lead times, while MPENet maintains competitive skill and shows advantages in capturing temporal evolution over longer periods. Cross-correlation analyses of MPENet and FourCastNet predictions with the HRRR data suggest that encoding critical cyclicity into the network architecture leads to improvements in the forecasting skill.

Artificial intelligence↗
NASA NTRSDate not supplied

Development of Physics-Based Transition Models for Unstructured-Mesh CFD Codes Using Deep Learning Models

Predicting transition locations over a vehicle surface is of fundamental importance for many engineering applications. With the transition information, the Reynolds-averaged Navier-Stokes (RANS) computations can turn on the turbulence model at the right locations so that drag, lift and other aerodynamic quantities can be accurately predicted. In contrast to the popularity of RANS-based transition modeling in which transition onset is governed by the turbulence equations, physics-based transition models that account for instability waves within the boundary layer, thus more compliant to flow physics, only gained more attention in recent years. This paper describes the development of a new physics-based transition model based on either the linear stability theory (LST) or parabolized stability equations (PSE). The model is designed to communicate with a structured or unstructured-mesh RANS solver back and forth in order to more accurately compute transition fronts over a three-dimensional body. In the developed model, the Python suite of interface codes in conjunction with the LASTRAC software can be executed autonomously to produce transition onset locations for a given laminar or RANS-computed transitional state. In addition, as a proof of concept, the tool set consists of a deep learning neural network model that has been designed and trained to predict instability wave evolutions inside the boundary layer for various instability wave mechanisms across a selected speed range. A machine-learned intelligent profile interpolation model has also been devised to enable reliable instability-wave spectra predictions with just a few points in the mean flow profiles.

Transition Modeling↗
DOE OSTI2024

Exploring the frontiers of condensed-phase chemistry with a general reactive machine learning potential

Abstract Atomistic simulation has a broad range of applications from drug design to materials discovery. Machine learning interatomic potentials (MLIPs) have become an efficient alternative to computationally expensive ab initio simulations. For this reason, chemistry and materials science would greatly benefit from a general reactive MLIP, that is, an MLIP that is applicable to a broad range of reactive chemistry without the need for refitting. Here we develop a general reactive MLIP (ANI-1xnr) through automated sampling of condensed-phase reactions. ANI-1xnr is then applied to study five distinct systems: carbon solid-phase nucleation, graphene ring formation from acetylene, biofuel additives, combustion of methane and the spontaneous formation of glycine from early earth small molecules. In all studies, ANI-1xnr closely matches experiment (when available) and/or previous studies using traditional model chemistry methods. As such, ANI-1xnr proves to be a highly general reactive MLIP for C, H, N and O elements in the condensed phase, enabling high-throughput in silico reactive chemistry experimentation.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2022

PlasmidHostFinder: Prediction of Plasmid Hosts Using Random Forest

Plasmids play a major role facilitating the spread of antimicrobial resistance between bacteria. Understanding the host range and dissemination trajectories of plasmids is critical for surveillance and prevention of antimicrobial resistance. Identification of plasmid host ranges could be improved using automated pattern detection methods compared to homology-based methods due to the diversity and genetic plasticity of plasmids. In this study, we developed a method for predicting the host range of plasmids using machine learning—specifically, random forests. We trained the models with 8,519 plasmids from 359 different bacterial species per taxonomic level; the models achieved Matthews correlation coefficients of 0.662 and 0.867 at the species and order levels, respectively. Our results suggest that despite the diverse nature and genetic plasticity of plasmids, our random forest model can accurately distinguish between plasmid hosts. This tool is available online through the Center for Genomic Epidemiology (https://cge.cbs.dtu.dk/services/PlasmidHostFinder/).

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2016

PHOTOMETRIC SUPERNOVA CLASSIFICATION WITH MACHINE LEARNING

Automated photometric supernova classification has become an active area of research in recent years in light of current and upcoming imaging surveys such as the Dark Energy Survey (DES) and the Large Synoptic Survey Telescope, given that spectroscopic confirmation of type for all supernovae discovered will be impossible. Here, we develop a multi-faceted classification pipeline, combining existing and new approaches. Our pipeline consists of two stages: extracting descriptive features from the light curves and classification using a machine learning algorithm. Our feature extraction methods vary from model-dependent techniques, namely SALT2 fits, to more independent techniques that fit parametric models to curves, to a completely model-independent wavelet approach. We cover a range of representative machine learning algorithms, including naive Bayes, k -nearest neighbors, support vector machines, artificial neural networks, and boosted decision trees (BDTs). We test the pipeline on simulated multi-band DES light curves from the Supernova Photometric Classification Challenge. Using the commonly used area under the curve (AUC) of the Receiver Operating Characteristic as a metric, we find that the SALT2 fits and the wavelet approach, with the BDTs algorithm, each achieve an AUC of 0.98, where 1 represents perfect classification. We find that a representative training set is essential for good classification, whatever the feature set or algorithm, with implications for spectroscopic follow-up. Importantly, we find that by using either the SALT2 or the wavelet feature sets with a BDT algorithm, accurate classification is possible purely from light curve data, without the need for any redshift information.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2012

Proactive Defense for Evolving Cyber Threats

There is great interest to develop proactive methods of cyber defense, in which future attack strategies are anticipated and these insights are incorporated into defense designs; however, little has been done to place this ambitious objective on a sound scientific foundation. Indeed, even fundamental issues associated with how the “arms race” between attackers and defenders actually leads to predictability in attacker activity, or how to effectively and scalably detect this predictability in the relational/temporal data streams generated by attacker/defender adaptation, haven’t been resolved. This LDRD project addressed many of these challenges and the results are briefly summarized here. We have characterized the predictability of attacker/defender coevolution and have leveraged our findings to create a framework for designing proactive defenses for large (organizational) networks. More specifically, this project applied rigorous predictability-based analytics to two central and complementary aspects of the network defense problem – attack strategies of the adversaries and vulnerabilities of the defenders’ systems – and used the results to develop a scientifically-grounded, practically-implementable methodology for designing proactive cyber defense systems. Briefly, predictive analysis of attack strategies involved first conducting predictability assessments to characterize attacker adaptation patterns in given domains, and then used these patterns to “train” adaptive defense systems capable of providing robust performance against both current and (near) future threats. The problem of identifying and prioritizing defender system vulnerabilities was addressed using statistical and machine learning to analyze a broad range of data (e.g., cyber, social media) on recently detected system vulnerabilities to “learn” classifiers that predict how likely it is that, and how soon, new vulnerabilities will be exploited. A variety of cyber threat case studies were developed and investigated throughout the project, one selected from the cyber security research community and one that is more comprehensive and of higher priority to SNL and to external national security partners. A sample of research results and application of this methodology are included in this report (as a series of peer-reviewed publications). For ease of reference the title and SAND number are included below.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2017

Every apple has a voice: using stable isotopes to teach about food sourcing and the water cycle

Abstract. Agricultural crops such as fruits take up irrigation and meteoric water and incorporate it into their tissue ( fruit water ) during growth, and the geographic origin of a fruit may be traced by comparing the H and O stable isotope composition ( δ 2 H and δ 18 O values) of fruit water to the global geospatial distribution of H and O stable isotopes in precipitation. This connection between common fruits and the global water cycle provides an access point to connect with a variety of demographic groups to educate about isotope hydrology and the water cycle. Within the context of a 1-day outreach activity designed for a wide spectrum of participants (high school students, undergraduate students, high school science teachers) we developed introductory lecture materials, in-class participatory demonstrations of fruit water isotopic measurement in real time, and a computer lab exercise to couple actual fruit water isotope data with open-source online geospatial analysis software. We assessed learning outcomes with pre- and post-tests tied to learning objectives, as well as participant feedback surveys. Results indicate that this outreach activity provided effective lessons on the basics of stable isotope hydrology and the water cycle. However, the computer lab exercise needs to be more specifically tailored to the abilities of each participant group. This pilot study provides a foundation for further development of outreach materials that can effectively engage a range of participant groups in learning about the water cycle and the ways in which humans modify the water cycle through agricultural activity.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2025

Model-free estimation of completeness, uncertainties, and outliers in atomistic machine learning using information theory

Abstract An accurate description of information is relevant for a range of problems in atomistic machine learning (ML), such as crafting training sets, performing uncertainty quantification (UQ), or extracting physical insights from large datasets. However, atomistic ML often relies on unsupervised learning or model predictions to analyze information contents from simulation or training data. Here, we introduce a theoretical framework that provides a rigorous, model-free tool to quantify information contents in atomistic simulations. We demonstrate that the information entropy of a distribution of atom-centered environments explains known heuristics in ML potential developments, from training set sizes to dataset optimality. Using this tool, we propose a model-free UQ method that reliably predicts epistemic uncertainty and detects out-of-distribution samples, including rare events in systems such as nucleation. This method provides a general tool for data-driven atomistic modeling and combines efforts in ML, simulations, and physical explainability.

36 MATERIALS SCIENCE↗
DOE OSTI2011

FY 2011 Miniature Spherical Retroreflectors (Final Report)

PNNL conducted a systematic optical modeling effort to study and compare retroreflector designs with respect to available optical materials. The optical Strehl ratio (SR) and cross section figures of merit were used to compare the performance of candidate optical designs. Further modeling was conducted to assess the relationships between standoff range and retroreflector diameter and cross section. Analysis showed that small diameter retroreflectors have very high optical link loss, since the net loss scales by retroreflector diameter to the 4th power. Optical retroreflection efficiency is also important, but extracting ideal performance from a complicated design must be tempered by the fact that the effective retroreflector range varies only as the fourth root of the cross section. It's therefore imperative to maintain a reasonable retroreflector diameter, but striving for perfect optical performance may be a costly effort that does not merit the modest performance gain. Several promising designs were developed for visible and mid-infrared applications, which embraced lessons learned in the standoff range study. These designs included a polymer clad glass core approach for visible wavelengths and a concentric shell design using chalcogenide glass for the mid-infrared. Injection molding feasibility studies demonstrated that it was practical to fabricate visible polymer retroreflectors using this approach. Such an approach may be a practical solution for the LLNL project. Our primary focus is to develop mid-infrared designs; therefore compression molding chalcogenide glass was adopted for fabricating mid-infrared retroreflectors. A significant effort was focused on developing practical compression molding methods that provided reasonable device yield. A custom compression molding press was developed in collaboration with a private engineering firm. Many successful molding studies were conducted during this effort, but ultimately the fragile nature of the chalcogenide glass and the unfavorable thermal expansion properties limited the feasibility of fabricating the thin shell structures needed for this design. An alternative Cat's Eye retroreflector design was therefore adopted. This design is constructed using two hemispherical structures that can be fabricated by conventional grinding and polishing methods or by molding with high yield. The cross section of this design is very good (and can be further improved by applying a highly reflective coating to the rear surface), but has reduced field of view as a design trade off as well as the isotropic nature of Luneburg-style retroreflector.

36 MATERIALS SCIENCE↗
DOE OSTI2016

SU-C-BRA-04: Automated Segmentation of Head-And-Neck CT Images for Radiotherapy Treatment Planning Via Multi-Atlas Machine Learning (MAML)

Purpose: Accurate image segmentation is a crucial step during image guided radiation therapy. This work proposes multi-atlas machine learning (MAML) algorithm for automated segmentation of head-and-neck CT images. Methods: As the first step, the algorithm utilizes normalized mutual information as similarity metric, affine registration combined with multiresolution B-Spline registration, and then fuses together using the label fusion strategy via Plastimatch. As the second step, the following feature selection strategy is proposed to extract five feature components from reference or atlas images: intensity (I), distance map (D), box (B), center of gravity (C) and stable point (S). The box feature B is novel. It describes a relative position from each point to minimum inscribed rectangle of ROI. The center-of-gravity feature C is the 3D Euclidean distance from a sample point to the ROI center of gravity, and then S is the distance of the sample point to the landmarks. Then, we adopt random forest (RF) in Scikit-learn, a Python module integrating a wide range of state-of-the-art machine learning algorithms as classifier. Different feature and atlas strategies are used for different ROIs for improved performance, such as multi-atlas strategy with reference box for brainstem, and single-atlas strategy with reference landmark for optic chiasm. Results: The algorithm was validated on a set of 33 CT images with manual contours using a leave-one-out cross-validation strategy. Dice similarity coefficients between manual contours and automated contours were calculated: the proposed MAML method had an improvement from 0.79 to 0.83 for brainstem and 0.11 to 0.52 for optic chiasm with respect to multi-atlas segmentation method (MA). Conclusion: A MAML method has been proposed for automated segmentation of head-and-neck CT images with improved performance. It provides the comparable result in brainstem and the improved result in optic chiasm compared with MA. Xuhua Ren and Hao Gao were partially supported by the NSFC (#11405105), the 973 Program (#2015CB856000), and the Shanghai Pujiang Talent Program (#14PJ1404500).

60 APPLIED LIFE SCIENCES↗
DOE OSTI2024

Performance on HPC Platforms Is Possible Without C++

Computing at large scales has become extremely challenging due to increasing heterogeneity in both hardware and software. More and more scientific workflows must tackle a range of scales and use machine learning and AI intertwined with more traditional numerical modeling methods, placing more demands on computational platforms. These constraints indicate a need to fundamentally rethink the way computational science is done and the tools that are needed to enable these complex workflows. The current set of C++-based solutions may not suffice, and relying exclusively upon C++ may not be the best option, especially because several newer languages and boutique solutions offer more robust design features to tackle the challenges of heterogeneity. In June 2023, we held a mini symposium that explored the use of newer languages and heterogeneity solutions that are not tied to C++ and that offer options beyond template metaprogramming and Parallel. For for performance and portability. In conclusion, we describe some of the presentations and discussion from the mini symposium in this article.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2014

TU-C-BRE-07: Quantifying the Clinical Impact of VMAT Delivery Errors Relative to Prior Patients’ Plans and Adjusted for Anatomical Differences

Purpose: There is increased interest in the Radiation Oncology Physics community regarding sensitivity of pre-treatment IMRT/VMAT QA to delivery errors. Consequently, tools mapping pre-treatment QA to the patient DVH have been developed. However, the quantity of plan degradation that is acceptable remains uncertain. Using DVHs adapted from prior patients’ plans, we developed a technique to determine the magnitude of various delivery errors required to degrade a treatment plan to outside the clinically accepted range. Methods: DVHs for relevant organs at risk were adapted from a population of prior patients’ plans using a machine learning algorithm to establish the clinically acceptable DVH range specific to the patient’s anatomy. We applied this technique to six low-risk prostate cancer patients treated with single-arc VMAT and compared error-induced DVH changes to the adapted DVHs to determine the magnitude of error required to push the plan outside of the acceptable range. The procedure follows: (1) Errors (systematic ' random shift of MLCs, gantry-MLC desynchronization, dose rate fluctuations, etc.) were simulated and degraded DVHs calculated using the Varian Eclipse TPS. (2) Adapted DVHs and acceptable ranges for DVHs were established. (3) Relevant dosimetric indices and corresponding acceptable ranges were calculated from the DVHs. Key indices included NTCP (Lyman-Kutcher-Burman Model) and QUANTEC’s dose-volume Objectives: s of V75Gy≤0.15 for the rectum and V75Gy≤0.25 for the bladder. Results: Degradations to the clinical plan became “unacceptable” for 19±29mm and 1.9±2.0mm systematic outward shifts of a single leaf and leaf bank, respectively. All other simulated errors fell within the acceptable range. Conclusion: Utilizing machine learning and prior patients’ plans one can predict a clinically acceptable range of DVH degradation for a specific patient. Comparing error-induced DVH degradations to this range, it is shown that single-arc VMAT plans for low-risk prostate are relatively insensitive to many potential delivery errors.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2021

Prognostics and Health Management in Nuclear Power Plants: An Updated Method-Centric Review With Special Focus on Data-Driven Methods

In a carbon-constrained world, future uses of nuclear power technologies can contribute to climate change mitigation as the installed electricity generating capacity and range of applications could be much greater and more diverse than with the current plants. To preserve the nuclear industry competitiveness in the global energy market, prognostics and health management (PHM) of plant assets is expected to be important for supporting and sustaining improvements in the economics associated with operating nuclear power plants (NPPs) while maintaining their high availability. Of interest are long-term operation of the legacy fleet to 80 years through subsequent license renewals and economic operation of new builds of either light water reactors or advanced reactor designs. Recent advances in data-driven analysis methods—largely represented by those in artificial intelligence and machine learning—have enhanced applications ranging from robust anomaly detection to automated control and autonomous operation of complex systems. The NPP equipment PHM is one area where the application of these algorithmic advances can significantly improve the ability to perform asset management. This paper provides an updated method-centric review of the full PHM suite in NPPs focusing on data-driven methods and advances since the last major survey article was published in 2015. The main approaches and the state of practice are described, including those for the tasks of data acquisition, condition monitoring, diagnostics, prognostics, and planning and decision-making. Research advances in non-nuclear power applications are also included to assess findings that may be applicable to the nuclear industry, along with the opportunities and challenges when adapting these developments to NPPs. Finally, this paper identifies key research needs in regard to data availability and quality, verification and validation, and uncertainty quantification.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2022

Dictionary Learning with Accumulator Neurons

The Locally Competitive Algorithm (LCA) uses local competition between non-spiking leaky integrator neurons to infer sparse representations, allowing for potentially real-time execution on massively parallel neuromorphic architectures such as Intel's Loihi processor. Here, we focus on the problem of inferring sparse representations from streaming video using dictionaries of spatiotemporal features optimized in an unsupervised manner for sparse reconstruction. Non-spiking LCA has previously been used to achieve unsupervised learning of spatiotemporal dictionaries composed of convolutional kernels from raw, unlabeled video. We demonstrate how unsupervised dictionary learning with spiking LCA (\hbox{S-LCA}) can be efficiently implemented using accumulator neurons, which combine a conventional leaky-integrate-and-fire (\hbox{LIF}) spike generator with an additional state variable that is used to minimize the difference between the integrated input and the spiking output. We demonstrate dictionary learning across a wide range of dynamical regimes, from graded to intermittent spiking, for inferring sparse representations of both static images drawn from the CIFAR database as well as video frames captured from a DVS camera. On a classification task that requires identification of the suite from a deck of cards being rapidly flipped through as viewed by a DVS camera, we find essentially no degradation in performance as the LCA model used to infer sparse spatiotemporal representations migrates from graded to spiking. We conclude that accumulator neurons are likely to provide a powerful enabling component of future neuromorphic hardware for implementing online unsupervised learning of spatiotemporal dictionaries optimized for sparse reconstruction of streaming video from event based DVS cameras.

artificial intelligence↗
DOE OSTI2025

Intercomparison of Deep Learning Model Architectures for Atmospheric River Prediction

With a rapid surge in the application of machine learning (ML) for a diverse range of tasks in climate science, the present study addresses a challenge for climate scientists when selecting the optimal ML or deep learning (DL) architecture for a given application. In particular, a DL intercomparison study was performed with a focus on forecasting the position of atmospheric rivers (ARs) on short-range time scales (up to 5-day lead times). AR predictions from multiple DL architectures, including various types of convolutional autoencoders and a vision transformer (ViT), were compared against ECMWF ERA5 reanalysis and hindcasts from a global climate model. DL models with similar trainable parameters were trained on ERA5 reanalysis data and AR positions derived from a thresholding algorithm to ensure a fair comparison among the DL models. Each model’s performance and accuracy in forecasting AR location and key input fields within a 5-day window were assessed using metrics of root-mean-square error, anomaly correlation, and mean intersection over union. The ViT architecture outperformed other autoencoder models in most of the metrics. Incorporating additional meteorological fields only yielded slight improvements in forecasting certain fields at longer lead times. The results also suggest that a smaller number of input time steps or smaller number of autoregressive steps can achieve better prediction skills, while also improving the overall computational efficiency. This research offers valuable insights into the strengths and weaknesses of different DL techniques for AR forecasting, hopefully guiding the development of improved models for forecasting this phenomenon.

54 ENVIRONMENTAL SCIENCES↗