Search NASA⌕ Search

SEARCH · Search NASA

Results for “classification algorithms”

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 1,315 records · Page 73
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

QoS-aware edge AI placement and scheduling with multiple implementations in FaaS-based edge computing

Resource constraints on the computing continuum require that we make smart decisions for serving AI-based services at the network edge. AI-based services typically have multiple implementations (e.g., image classification implementations include SqueezeNet, DenseNet, and others) with varying trade-offs (e.g., latency and accuracy). The question then is how should AI-based services be placed across Function-as-a-Service (FaaS) based edge computing systems in order to maximize total Quality-of-Service (QoS). To address this question, we propose a problem that jointly aims to solve (i) edge AI service placement and (ii) request scheduling. These are done across two time-scales (one for placement and one for scheduling). Here we first cast the problem as an integer linear program. We then decompose the problem into separate placement and scheduling subproblems and prove that both are NP-hard. We then propose a novel placement algorithm that places services while considering device-to-device communication across edge clouds to offload requests to one another. Our results show that the proposed placement algorithm is able to outperform a state-of-the-art placement algorithm for AI-based services, and other baseline heuristics, with regard to maximizing total QoS. Additionally, we present a federated learning-based framework, FLIES, to predict the future incoming service requests and their QoS requirements. Our results also show that our FLIES algorithm is able to outperform a standard decentralized learning baseline for predicting incoming requests and show comparable predictive performance when compared to centralized training.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1995

Searching for patterns in remote sensing image databases using neural networks

We have investigated a method, based on a successful neural network multispectral image classification system, of searching for single patterns in remote sensing databases. While defining the pattern to search for and the feature to be used for that search (spectral, spatial, temporal, etc.) is challenging, a more difficult task is selecting competing patterns to train against the desired pattern. Schemes for competing pattern selection, including random selection and human interpreted selection, are discussed in the context of an example detection of dense urban areas in Landsat Thematic Mapper imagery. When applying the search to multiple images, a simple normalization method can alleviate the problem of inconsistent image calibration. Another potential problem, that of highly compressed data, was found to have a minimal effect on the ability to detect the desired pattern. The neural network algorithm has been implemented using the PVM (Parallel Virtual Machine) library and nearly-optimal speedups have been obtained that help alleviate the long process of searching through imagery.

Paola, Justin D.↗
NASA NTRS2014

Assimilation of Freeze - Thaw Observations into the NASA Catchment Land Surface Model

The land surface freeze-thaw (F-T) state plays a key role in the hydrological and carbon cycles and thus affects water and energy exchanges and vegetation productivity at the land surface. In this study, we developed an F-T assimilation algorithm for the NASA Goddard Earth Observing System, version 5 (GEOS-5) modeling and assimilation framework. The algorithm includes a newly developed observation operator that diagnoses the landscape F-T state in the GEOS-5 Catchment land surface model. The F-T analysis is a rule-based approach that adjusts Catchment model state variables in response to binary F-T observations, while also considering forecast and observation errors. A regional observing system simulation experiment was conducted using synthetically generated F-T observations. The assimilation of perfect (error-free) F-T observations reduced the root-mean-square errors (RMSE) of surface temperature and soil temperature by 0.206 C and 0.061 C, respectively, when compared to model estimates (equivalent to a relative RMSE reduction of 6.7 percent and 3.1 percent, respectively). For a maximum classification error (CEmax) of 10 percent in the synthetic F-T observations, the F-T assimilation reduced the RMSE of surface temperature and soil temperature by 0.178 C and 0.036 C, respectively. For CEmax=20 percent, the F-T assimilation still reduces the RMSE of model surface temperature estimates by 0.149 C but yields no improvement over the model soil temperature estimates. The F-T assimilation scheme is being developed to exploit planned operational F-T products from the NASA Soil Moisture Active Passive (SMAP) mission.

GEOS5↗
NASA NTRS2011

CropEx Web-Based Agricultural Monitoring and Decision Support

CropEx is a Web-based agricultural Decision Support System (DSS) that monitors changes in crop health over time. It is designed to be used by a wide range of both public and private organizations, including individual producers and regional government offices with a vested interest in tracking vegetation health. The database and data management system automatically retrieve and ingest data for the area of interest. Another stores results of the processing and supports the DSS. The processing engine will allow server-side analysis of imagery with support for image sub-setting and a set of core raster operations for image classification, creation of vegetation indices, and change detection. The system includes the Web-based (CropEx) interface, data ingestion system, server-side processing engine, and a database processing engine. It contains a Web-based interface that has multi-tiered security profiles for multiple users. The interface provides the ability to identify areas of interest to specific users, user profiles, and methods of processing and data types for selected or created areas of interest. A compilation of programs is used to ingest available data into the system, classify that data, profile that data for quality, and make data available for the processing engine immediately upon the data s availability to the system (near real time). The processing engine consists of methods and algorithms used to process the data in a real-time fashion without copying, storing, or moving the raw data. The engine makes results available to the database processing engine for storage and further manipulation. The database processing engine ingests data from the image processing engine, distills those results into numerical indices, and stores each index for an area of interest. This process happens each time new data is ingested and processed for the area of interest, and upon subsequent database entries, the database processing engine qualifies each value for each area of interest and conducts a logical processing of results indicating when and where thresholds are exceeded. Reports are provided at regular, operator-determined intervals that include variances from thresholds and links to view raw data for verification, if necessary. The technology and method of development allow the code base to easily be modified for varied use in the real-time and near-real-time processing environments. In addition, the final product will be demonstrated as a means for rapid draft assessment of imagery.

Harvey. Craig↗
DOE OSTI2019

Probabilistic variable-length segmentation of protein sequences for discriminative motif discovery (DiMotif) and sequence embedding (ProtVecX)

In this paper, we present peptide-pair encoding (PPE), a general-purpose probabilistic segmentation of protein sequences into commonly occurring variable-length sub-sequences. The idea of PPE segmentation is inspired by the byte-pair encoding (BPE) text compression algorithm, which has recently gained popularity in subword neural machine translation. We modify this algorithm by adding a sampling framework allowing for multiple ways of segmenting a sequence. PPE segmentation steps can be learned over a large set of protein sequences (Swiss-Prot) or even a domain-specific dataset and then applied to a set of unseen sequences. This representation can be widely used as the input to any downstream machine learning tasks in protein bioinformatics. In particular, here, we introduce this representation through protein motif discovery and protein sequence embedding. (i) DiMotif: we present DiMotif as an alignment-free discriminative motif discovery method and evaluate the method for finding protein motifs in three different settings: (1) comparison of DiMotif with two existing approaches on 20 distinct motif discovery problems which are experimentally verified, (2) classification-based approach for the motifs extracted for integrins, integrin-binding proteins, and biofilm formation, and (3) in sequence pattern searching for nuclear localization signal. The DiMotif, in general, obtained high recall scores, while having a comparable F1 score with other methods in the discovery of experimentally verified motifs. Having high recall suggests that the DiMotif can be used for short-list creation for further experimental investigations on motifs. In the classification-based evaluation, the extracted motifs could reliably detect the integrins, integrin-binding, and biofilm formation-related proteins on a reserved set of sequences with high F1 scores. (ii) ProtVecX: we extend k-mer based protein vector (ProtVec) embedding to variablelength protein embedding using PPE sub-sequences. We show that the new method of embedding can marginally outperform ProtVec in enzyme prediction as well as toxin prediction tasks. In addition, we conclude that the embeddings are beneficial in protein classification tasks when they are combined with raw amino acids k-mer features.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2018

Probabilistic variable-length segmentation of protein sequences for discriminative motif discovery (DiMotif) and sequence embedding (ProtVecX)

ABSTRACT In this paper, we present peptide-pair encoding (PPE), a general-purpose probabilistic segmentation of protein sequences into commonly occurring variable-length sub-sequences. The idea of PPE segmentation is inspired by the byte-pair encoding (BPE) text compression algorithm, which has recently gained popularity in subword neural machine translation. We modify this algorithm by adding a sampling framework allowing for multiple ways of segmenting a sequence. PPE segmentation steps can be learned over a large set of protein sequences (Swiss-Prot) or even a domain-specific dataset and then applied to a set of unseen sequences. This representation can be widely used as the input to any downstream machine learning tasks in protein bioinformatics. In particular, here, we introduce this representation through protein motif discovery and protein sequence embedding. (i) DiMotif: we present DiMotif as an alignment-free discriminative motif discovery method and evaluate the method for finding protein motifs in three different settings: (1) comparison of DiMotif with two existing approaches on 20 distinct motif discovery problems which are experimentally verified, (2) classification-based approach for the motifs extracted for integrins, integrin-binding proteins, and biofilm formation, and (3) in sequence pattern searching for nuclear localization signal. The DiMotif, in general, obtained high recall scores, while having a comparable F1 score with other methods in the discovery of experimentally verified motifs. Having high recall suggests that the DiMotif can be used for short-list creation for further experimental investigations on motifs. In the classification-based evaluation, the extracted motifs could reliably detect the integrins, integrin-binding, and biofilm formation-related proteins on a reserved set of sequences with high F1 scores. (ii) ProtVecX: we extend k-mer based protein vector (ProtVec) embedding to variable-length protein embedding using PPE sub-sequences. We show that the new method of embedding can marginally outperform ProtVec in enzyme prediction as well as toxin prediction tasks. In addition, we conclude that the embeddings are beneficial in protein classification tasks when they are combined with raw k-mer features. Availability Implementations of our method will be available under the Apache 2 licence at http://llp.berkeley.edu/dimotif and http://llp.berkeley.edu/protvecx .

Asgari, Ehsaneddin↗
NASA NTRS2001

Resolution Enhancement of Spaceborne Radiometer Images

Our progress over the last year has been along several dimensions: 1. Exploration and understanding of Earth Observatory System (EOS) mission with available data from NASA. 2. Comprehensive review of state of the art techniques and uncovering of limitations to be investigated (e.g. computational, algorithmic ...). and 3. Preliminary development of resolution enhancement algorithms. With the advent of well-collaborated satellite microwave radiometers, it is now possible to obtain long time series of geophysical parameters that are important for studying the global hydrologic cycle and earth radiation budget. Over the world's ocean, these radiometers simultaneously measure profiles of air temperature and the three phases of atmospheric water (vapor, liquid, and ice). In addition, surface parameters such as the near surface wind speed, the sea surface temperature, and the sea ice type and concentration can be retrieved. The special sensor microwaves imager SSM/I has wide application in atmospheric remote sensing over the ocean and provide essential inputs to numerical weather-prediction models. SSM/I data has also been used for land and ice studies, including snow cover classification measurements of soil and plant moisture contents, atmospheric moisture over land, land surface temperature and mapping polar ice. The brightness temperature observed by SSM/I is function of the effective brightness temperature of the earth's surface and the emission scattering and attenuation of the atmosphere. Advanced Microwave Scanning Radiometer (AMSR) is a new instrument that will measure the earth radiation over the spectral range from 7 to 90 GHz. Over the world's ocean, it will be possible to retrieve the four important geographical parameters SST, wind speed, vertically integrated water vapor, vertically integrated cloud liquid water L.

Krim, Hamid↗
NASA NTRSDate not supplied

Expanding NeMO-Net Machine Learning Capabilities for Citizen Science

NASA NeMO-Net, the neural multi-modal observation and training network for global coral reef assessment, is an open-source deep convolutional neural network and interactive active learning training software aiming to accurately assess the present and past dynamics of coral reef ecosystems through determination of percent living cover and morphology as well as mapping of spatial distribution. We present an interactive citizen science video game, released this April, for desktop and iOS devices where users interactively label morphology classifications over mm-scale 3D coral reef imagery captured using diver photomosaic imagery, the UAV enabled NASA FluidCam instrument, and satellite datasets. To date, the application has had over 40,000 downloads and over 60,000 unique coral reef classifications, each filtered through a user-based rating and expert evaluation system. We also present results from NeMO-Net’s convolutional neural network (CNN) models used to semantically segment 2D satellite imagery as well as projections of 3D coral reconstructions using user input data as training datasets. Fusing datasets using machine learning from multiple remote sensing platforms presents novel methodologies for assessing the health of coral ecosystems, which are critically endangered by a changing climate. In partnering with Mission Blue, the National Oceanic and Atmospheric Administration (NOAA), and the Living Oceans Foundation (LOF), NeMO-Net leverages an international consortium of subject matter experts to provide both proper training for citizen scientists and the generation of a labeled datasets to ingest into machine learning algorithms for global coral reef identification.

NeMO-Net↗
NASA NTRSDate not supplied

Expanding NeMO-Net Machine Learning Capabilities for Citizen Science

NASA NeMO-Net, the neural multi-modal observation and training network for global coral reef assessment, is an open-source deep convolutional neural network and interactive active learning training software aiming to accurately assess the present and past dynamics of coral reef ecosystems through determination of percent living cover and morphology as well as mapping of spatial distribution. We present an interactive citizen science video game, released this April, for desktop and iOS devices where users interactively label morphology classifications over mm-scale 3D coral reef imagery captured using diver photomosaic imagery, the UAV enabled NASA FluidCam instrument, and satellite datasets. To date, the application has had over 40,000 downloads and over60,000 unique coral reef classifications, each filtered through a user-based rating and expert evaluation system. We also present results from NeMO-Net’s convolutional neural network (CNN) models used to semantically segment 2D satellite imagery as well as projections of 3D coral reconstructions using user input data as training datasets. Fusing datasets using machine learning from multiple remote sensing platforms presents novel methodologies for assessing the health of coral ecosystems, which are critically endangered by a changing climate. In partnering with Mission Blue, the National Oceanic and Atmospheric Administration (NOAA), and the Living Oceans Foundation (LOF), NeMO-Net leverages an international consortium of subject matter experts to provide both proper training for citizen scientists and the generation of a labeled datasets to ingest into machine learning algorithms for global coral reef identification.

NeMO-Net↗
DOE OSTI2024

ZMPY3D: accelerating protein structure volume analysis through vectorized 3D Zernike moments and Python-based GPU integration

Abstract Motivation Volumetric 3D object analyses are being applied in research fields such as structural bioinformatics, biophysics, and structural biology, with potential integration of artificial intelligence/machine learning (AI/ML) techniques. One such method, 3D Zernike moments, has proven valuable in analyzing protein structures (e.g., protein fold classification, protein–protein interaction analysis, and molecular dynamics simulations). Their compactness and efficiency make them amenable to large-scale analyses. Established methods for deriving 3D Zernike moments, however, can be inefficient, particularly when higher order terms are required, hindering broader applications. As the volume of experimental and computationally-predicted protein structure information continues to increase, structural biology has become a “big data” science requiring more efficient analysis tools. Results This application note presents a Python-based software package, ZMPY3D, to accelerate computation of 3D Zernike moments by vectorizing the mathematical formulae and using graphical processing units (GPUs). The package offers popular GPU-supported libraries such as CuPy and TensorFlow together with NumPy implementations, aiming to improve computational efficiency, adaptability, and flexibility in future algorithm development. The ZMPY3D package can be installed via PyPI, and the source code is available from GitHub. Volumetric-based protein 3D structural similarity scores and transform matrix of superposition functionalities have both been implemented, creating a powerful computational tool that will allow the research community to amalgamate 3D Zernike moments with existing AI/ML tools, to advance research and education in protein structure bioinformatics. Availability and implementation ZMPY3D, implemented in Python, is available on GitHub (https://github.com/tawssie/ZMPY3D) and PyPI, released under the GPL License.

Lai, Jhih-Siang (ORCID:0000000156775890)↗
NASA NTRS2011

Methodology and Results of the Near-Earth Object (NEO) Human Space Flight (HSF) Accessible Targets Study (NHATS)

Near-Earth Asteroids (NEAs) have been identified by the current administration as potential destinations for human explorers during the mid-2020s. While the close proximity of these objects' orbits to Earth's orbit creates a risk of highly damaging or catastrophic impacts, it also makes some of these objects particularly accessible to spacecraft departing Earth, and this presents unique opportunities for solar system science and humanity's first ventures beyond cislunar space. Planning such ambitious missions first requires the selection of potentially accessible targets from the growing population of nearly 7,800 NEAs. To accomplish this, NASA is conducting the Near-Earth Object (NEO) Human Space Flight (HSF) Accessible Targets Study (NHATS). Phase I of the NHATS was executed during September of 2010, and Phase II was completed by early March of 2011. The study is ongoing because previously undetected NEAs are being discovered constantly, which has motivated an effort to automate the analysis algorithms in order to provide continuous monitoring of NEA accessibility. The NHATS analysis process consists of a trajectory filter and a minimum maximum estimated size criterion. The trajectory filter employs the method of embedded trajectory grids to compute all possible ballistic round-trip mission trajectories to every NEA in the Jet Propulsion Laboratory (JPL) Small-Body Database (SBDB) and stores all solutions that satisfy the trajectory filter criteria. An NEA must offer at least one qualifying trajectory solution to pass the trajectory filter. The Phase II NHATS filter criteria were purposely chosen to be highly inclusive, requiring Earth departure date between January 1st, 2015 and December 31st, 2040, total round-trip flight time <= 450 days, stay time at the NEA >= 8 days, Earth departure C(sub 3) energy <= 60 km(exp 2)/s(exp 2), total mission delta-v <= 12 km/s (including an Earth departure maneuver from a 400 km altitude circular parking orbit), and a maximum atmospheric re-entry speed of 12 km/s. After determining which NEAs offer at least one trajectory solution meeting the criteria, the estimated size constraint is then imposed whereby those NEAs may only be considered NHATS- qualifying NEAs if their maximum estimated size is >= 30 m. This corresponds to an absolute magnitude H <= 26.5 with an assumed albedo p = 0:05. The following is a brief high-level summary of the Phase II study results. Of the 7,665 NEAs in the SBDB as of February 3rd, 2011, 765 NEAs passed the trajectory filter and yielded a total of 79,157,604 trajectory solutions. The trajectory solutions for each NEA are post-processed into Pork Chop Contour (PCC) plots which show total mission delta-v as a function of Earth departure date and total mission duration. Although the PCC plots necessarily compress a very multi-dimensional design space into a two-dimensional plot, they permit rapid assessment of the breadth and quality of an NEA's available Earth departure season and clearly indicate the regions of the trajectory design space which warrant further analysis and optimization. The PCC plot for the NEA with the greatest number of NHATS-qualifying trajectory solutions, 2000 SG-344, is shown. Of the 765 NEAs which passed the Phase II trajectory filter, a total of 590 NEAs also satisfied the further constraint of maximum estimated size >= 30 m. The distributions of osculating heliocentric orbital semi-major axis (a), eccentricity (e), and inclination (i), for those 590 NEAs are shown. Note that the semi-latus rectum used is equal to alpha (1-e(exp 2)). To further our understanding of round-trip trajectory accessibility dynamics, it is instructive to examine the distribution of the NHATS-Qualifying NEAs according to orbit classification. NEAs are grouped into four orbit families: Atiras (aphelion < 0.983 AU), Atens (aphelion > 0.983 AU, alpha < 1.0 AU), Apollos (perihelion < 1.017 AU, alpha > 1.0 AU), and Amors (1.017 < perihelion < 1.3 AU). Of the 765 NEAhich satisfied the NHATS trajectory criteria, none are Atiras, 193 are Atens (31% of known Atens), 456 are Apollos (11% of known Apollos), and 116 are Amors (4% of known Amors). While Apollos comprise 60% of the NEAs which pass the NHATS trajectory filter and Atens comprise only 25%, the percentages according to orbit family are perhaps more relevant. Note that only 11% of known Apollos passed the trajectory filter while 31% of known Atens passed. These simple statistics alone strongly suggest that Aten orbits possess features which tend to enhance their round-trip trajectory accessibility as compared to Apollos or Amors. This is significant because Atens' orbits cause them to spend considerable time in Earth's daytime sky, making them difficult to discover and track using ground-based observing assets. In this paper we will detail the NHATS analysis algorithms, present and analyze all NHATS results to date, and discuss aspects of HSF mission architecture design for future NEA missions.

Barbee, Brent↗
DOE OSTI2022

$\mathrm{SIM}$-$\mathrm{STEM}$ $\mathrm{Lab}$: Incorporating Compressed Sensing Theory for Fast STEM Simulation

We report recently it has been shown that precise dose control and an increase in the overall acquisition speed of atomic resolution scanning transmission electron microscope (STEM) images can be achieved by acquiring only a small fraction of the pixels in the image experimentally and then reconstructing the full image using an inpainting algorithm. In this paper, we apply the same inpainting approach (a form of compressed sensing) to simulated, sub-sampled atomic resolution STEM images. We find that it is possible to significantly sub-sample the area that is simulated, the number of g-vectors contributing the image, and the number of frozen phonon configurations contributing to the final image while still producing an acceptable fit to a fully sampled simulation. Here we discuss the parameters that we use and how the resulting simulations can be quantifiably compared to the full simulations. As with any Compressed Sensing methodology, care must be taken to ensure that isolated events are not excluded from the process, but the observed increase in simulation speed provides significant opportunities for real time simulations, image classification and analytics to be performed as a supplement to experiments on a microscope to be developed in the future

47 OTHER INSTRUMENTATION↗
NASA NTRS2000

Satellite Data Sets in the Polar Regions

We have generated about two decades of consistently derived geophysical parameters in the polar regions. The key parameters are sea ice concentration, surface temperature, albedo, and cloud cover statistics. Sea ice concentrations were derived from the Scanning Multichannel Microwave Radiometer (SMMR) data and the Special Scanning Cl Microwave Imager (SSM/I) data from several platforms using the enhanced Bootstrap Algorithm for the period 1978 through 1999. The new algorithm reduces the errors associated with spatial and temporal variations in the emissivity and surface temperatures of sea ice. Also, bad data at ocean/land interfaces are identified and deleted in an unsupervised manner. Surface ice temperature, albedo and cloud cover statistics are derived simultaneously from the Advanced Very High Resolution Radiometer (AVHRR) data from 1981 through 1999 and mapped at a higher resolution but the same format as the ice concentration data. The technique makes use these co-registered ice concentration maps to enable cloud masking to be done separately for open ocean, sea ice and land areas. The effect of inversion is minimized by taking into consideration the expected changes in the effect of inversion with altitude, especially in the Antarctic. A technique for ice type regional classification has also been developed using multichannel cluster analysis and a neural network. This provide a means to identify large areas of thin ice, first year ice, and older ice types. The data sets have been shown to be coherent with each other and provide a powerful tool for in depth studies of the currently changing Arctic and Antarctic environment.

Comiso, Josefino C.↗
DOE OSTI2022

A phenology- and trend-based approach for accurate mapping of sea-level driven coastal forest retreat

The rapid replacement of upland forest by encroaching marshland is a striking manifestation of global sea-level rise (SLR). Timely and high-resolution information on the location and extent of transition forest (the ecotone between upland forest and marsh where tree mortality due to seawater intrusion begins) is fundamental to understanding the processes and patterns of SLR-driven landscape reorganization. Despite its significance, accurate characterization of salt-impacted transition forest remains challenging due to the complexity of coastal environments, scarcity of ground-truth data, and the lack of effective mapping algorithms. Here we use the full archive of Landsat images between 1984 and 2021 to investigate the spectral, temporal, and phenological characteristics of transition forest, and develop a robust framework for monitoring coastal vegetation shifts in the mid-Atlantic U.S., a global SLR hotspot. Here, we found that transition forest exhibits strong negative NDVI trends and a deviation of land surface phenology from marsh and upland forest that distinguishes itself from surrounding vegetation. By integrating temporal trends and land surface phenology, our results demonstrate superior discrimination between marsh and coastal forests to existing map products (e.g. NOAA Coastal Change Analysis Program, National Land Cover Database) that allows a reliable identification of the coastal treeline. We applied the approach to map regional land cover in 1985, 2000 and 2020 (overall classification accuracy >92%) and found that the area of coastal forest decreased by 22.0% from 1985 to 2020, the majority of which transitioned to marshland (92.3%, 5.3 × 10 3 ha). Based upon fine-scale patterns of coastal transgression, we created a practical workflow for spatially explicit quantification of forest retreat rates. Concurrent with rising sea level, coastal forests migrated upslope from 0.63 (± 0.27) m above sea level in 1985 to 0.78 (± 0.32) m above sea level in 2020, and horizontal forest retreat rates accelerated from 3.1 (range of 0–36) m yr -1 during 1985–2000 to 4.7 (0–55) m yr -1 during 2001–2020. As SLR continues to accelerate, our study may serve as a scalable solution for consistent tracking of coastal landscape evolution that is urgently needed for sustainable forest and wetland management.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2020

Cognition at the Point of Sensing

Over the last 15 years, compressive sensing techniques have been developed which have the potential to greatly reduce the amount of data collected by systems while preserving the amount of information obtained. A cost of this efficiency is that a computationally-intensive optimization routine must be used to put the sensed data into a form that a person can interpret. At the same time, machine learning techniques have experienced tremendous growth as well. Machines have demonstrated the ability learn how to effectively perform tasks such as detection and classification at speeds much faster than humanly possible. Our goal in this project was to study the feasibility of using compressive sensing systems "at the edge." That is, how can compressive sensing sensors be deployed such that information is created at the remote sensor rather than sending raw data to a central processing location? Studies were performed to analyze whether machine learning could be done on the compressively sensed data in its raw form. If a machine is performing the task, is it possible to do so without putting the data into a human interpretable form? We show that this is possible for some systems, in particular a compressive sensing snapshot imaging spectrometer. Machine learning tasks were demonstrated to be more effective and more robust to noise when the machine learning algorithm worked on data in its raw form. This system is shown to outperform a traditional spectrometer. Techniques for reducing the complexity of the reconstruction routine were also analyzed. Techniques for such as data regularization, deep neural networks, and matrix completion were studied and shown to have benefits over traditional reconstruction techniques. In this project we showed that compressive sensing sensors are indeed feasible at the edge. As always, sensors and algorithms must be carefully tuned to work in the constrained environment. In this project we developed tools and techniques to enable those analyses.

47 OTHER INSTRUMENTATION↗
DOE OSTI2022

Demonstration of Decentralized Physics-Driven Learning

In typical artificial neural networks, neurons adjust according to global calculations of a central processor, but in the brain, neurons and synapses self-adjust based on local information. Contrastive learning algorithms have recently been proposed to train physical systems, such as fluidic, mechanical, or electrical networks, to perform machine-learning tasks from local evolution rules. However, to date, such systems have only been implemented in silico due to the engineering challenge of creating elements that autonomously evolve based on their own response to two sets of global boundary conditions. Here, we introduce and implement a physics-driven contrastive learning scheme for a network of variable resistors, using circuitry to locally compare the response of two identical networks subjected to the two different sets of boundary conditions. Using this method, our system effectively trains itself, optimizing its resistance values without the use of a central processor or external information storage. Once the system is trained for a specified allostery, regression, or classification task, the task is subsequently performed rapidly and automatically by the physical imperative to minimize power dissipation in response to the given voltage inputs. Here, we demonstrate that, unlike typical computers, such learning systems are robust to extreme damage (and thus manufacturing defects) due to their decentralized learning. Our twin-network approach is therefore readily scalable to extremely large or nonlinear networks, where its distributed nature will be an enormous advantage; a laboratory network of only 500 edges will already outpace its in silico counterpart.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
NASA NTRS1989

Active microwave classification of sea ice

Radar backscatter studies of Arctic sea ice have been carried out over a number of years with the intent to acquire physical property information through the examination of microwave signatures. The breadth of these studies continues to expand; as an example, measurements are now conducted at frequencies from 500 MHz to about 100 GHz. One of the scientific goals of this work has been to develop an improved outstanding of the scattering processes at play. A second, equally important goal has been to apply the knowledge gained in examining the backscatter response of ice and snow made in conjunction with the detailed scene characterizations, the insight gained through theoretical modeling and parametric study, and the data entered into the radar signature library to develop procedures to convert microwave signal information (available in the very near future) into valuable data products. This should ultimately provide a better understanding of the environment. The author discusses what has been learned through the many efforts associated with the near-surface scatterometer measurement programs and how the knowledge gained is assisting in the development of future sea ice type satellite algorithms. The logic and mechanisms used in discriminating sea ice types are presented.

Onstott, Robert G.↗
NASA NTRS2019

The Cumulus and Stratocumulus CloudSat-CALIPSO Dataset (CASCCAD)

Low clouds continue to contribute greatly to the uncertainty in cloud feedback estimates. Depending on whether a region is dominated by cumulus (Cu) or stratocumulus (Sc) clouds, the interannual low-cloud feedback is somewhat different in both spaceborne and large-eddy simulation studies. Therefore, simulating the correct amount and variation of the Cu and Sc cloud distributions could be crucial to predict future cloud feedbacks. Here we document spatial distributions and profiles of Sc and Cu clouds derived from Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) and CloudSat measurements. For this purpose, we create a new dataset called the Cumulus And Stratocumulus CloudSat-CALIPSO Dataset (CASCCAD), which identifies Sc, broken Sc, Cu under Sc, Cu with stratiform outflow and Cu. To separate the Cu from Sc, we design an original method based on the cloud height, horizontal extent, vertical variability and horizontal continuity, which is separately applied to both CALIPSO and combined CloudSat–CALIPSO observations. First, the choice of parameters used in the discrimination algorithm is investigated and validated in selected Cu, Sc and Sc–Cu transition case studies. Then, the global statistics are compared against those from existing passive- and active-sensor satellite observations. Our results indicate that the cloud optical thickness – as used in passive-sensor observations – is not a sufficient parameter to discriminate Cu from Sc clouds, in agreement with previous literature. Using clustering-derived datasets shows better results although one cannot completely separate cloud types with such an approach. On the contrary, classifying Cu and Sc clouds and the transition between them based on their geometrical shape and spatial heterogeneity leads to spatial distributions consistent with prior knowledge of these clouds, from ground-based, ship-based and field campaigns. Furthermore, we show that our method improves existing Sc–Cu classifications by using additional information on cloud height and vertical cloud fraction variation. Finally, the CASCCAD datasets provide a basis to evaluate shallow convection and stratocumulus clouds on a global scale in climate models and potentially improve our understanding of low-level cloud feedbacks. The CASCCAD dataset (Cesana, 2019, https://doi.org/10.5281/zenodo.2667637) is available on the Goddard Institute for Space Studies (GISS) website at https://data.giss.nasa.gov/clouds/casccad/ (last access: 5 November 2019) and on the zenodo website at https://zenodo.org/record/2667637 (last access: 5 November 2019).

Cesana, Gregory V.↗