Search NASA⌕ Search

SEARCH · Search NASA

Results for “misclassification”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 37 records · Page 2
All sourcesNASA NTRSDOE OSTI
NASA NTRS1990

Experimental fault characterization of a neural network

The effects of a variety of faults on a neural network is quantified via simulation. The neural network consists of a single-layered clustering network and a three-layered classification network. The percentage of vectors mistagged by the clustering network, the percentage of vectors misclassified by the classification network, the time taken for the network to stabilize, and the output values are all measured. The results show that both transient and permanent faults have a significant impact on the performance of the measured network. The corresponding mistag and misclassification percentages are typically within 5 to 10 percent of each other. The average mistag percentage and the average misclassification percentage are both about 25 percent. After relearning, the percentage of misclassifications is reduced to 9 percent. In addition, transient faults are found to cause the network to be increasingly unstable as the duration of a transient is increased. The impact of link faults is relatively insignificant in comparison with node faults (1 versus 19 percent misclassified after relearning). There is a linear increase in the mistag and misclassification percentages with decreasing hardware redundancy. In addition, the mistag and misclassification percentages linearly decrease with increasing network size.

Tan, Chang-Huong↗
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 OSTI2021

Predicting nepheline precipitation in waste glasses using ternary submixture model and machine learning

Nepheline precipitation in nuclear waste glasses during vitrification can be detrimental due to its negative effect on chemical durability. Developing models to accurately predict nepheline precipitation from compositions is important to increase waste loading since existing models can be overly conservative. In this study, an expanded dataset containing 955 glasses was compiled from literature data, where 355 glasses are for high-level waste (HLW). Previously developed submixture models were refitted using the new dataset, where a misclassification rate of 7.8% was achieved. Nine machine learning (ML) algorithms (e.g., k-nearest neighbor, Gaussian process regression, artificial neural network, support vector machine, decision tree, etc.) were applied to evaluate their ability of predicting nepheline precipitation from compositions. Model accuracy, precision, recall/sensitivity, and F1 score were systemically compared between different ML algorithms and modeling protocols. Good model prediction with an accuracy ~0.9 (misclassification rate of ~10%) was observed with different algorithms under certain protocol. This study evaluated various ML models to predict nepheline precipitations in waste glasses, highlighting the importance of data preparation, modeling protocol, and their effect on model stability and reproducibility. The results provide insights into applying ML to predict glass properties and suggest areas for future research on modeling nepheline precipitations.

Lu, Xiaonan↗
DOE OSTI2018

Discrimination of Anthropogenic Events and Tectonic Earthquakes in Utah Using a Quadratic Discriminant Function Approach with Local Distance Amplitude Ratios

Most of the commonly used seismic discrimination approaches are designed for teleseismic and regional data. To monitor for the smallest events, some of these discriminants have been adapted for local distances (< 200 km), with mixed level of success. For this, we take advantage of the variety of seismic sources, including nontraditionally studied anthropogenic sources and the existence of a dense regional seismic network in the Utah region to evaluate amplitude ratio seismic discrimination at local distances. First, we explored phase-amplitude Pg-to-Sg ratios for multiple frequency bands to classify events in a dataset that comprises populations of single-shot surface explosions, shallow and deep ripple-fired mining blasts, mining-induced events (MIEs), and tectonic earthquakes. We achieved a success rate of about 59%–83%. Then, for the same dataset, we combined the Pg-to-Sg phase-amplitude ratios with Sg-to-Rg spectral amplitude ratios in a multivariate quadratic discriminant function (QDF) approach. For two-category pairwise classification, seven of ten population pairs show misclassification rates of about 20% or less, with five pairs showing rates of about 10% or less. The approach performs best for the pair involving the populations of single-shot explosions and MIEs. By combining both Pg-to-Sg and Rg-to-Sg ratios in the multivariate QDFs, we are able to achieve an average improvement of about 4%–14% in misclassification rates compared with Pg-to-Sg ratios alone. When all five event populations are considered simultaneously, as expected, the potential for misclassification increases, and our QDF approach using both Pg-to-Sg and Rg-to-Sg ratios achieves an average success rate of about 74% compared with the rate of about 86% for two-category pairwise classification.

58 GEOSCIENCES↗
NASA NTRS1975

Study of recreational land and open space using Skylab imagery

The author has identified the following significant results. An analysis of the statistical uniqueness of each of the signatures of the Gratiot-Saginaw State Game Area was made by computing a matrix of probabilities of misclassification for all possible signature pairs. Within each data set, the 35 signatures were then aggregated into a smaller set of composite signatures by combining groups of signatures having high probabilities of misclassification. Computer separation of forest denisty classes was poor with multispectral scanner data collected on 5 August 1973. Signatures from the scanner data were further analyzed to determine the ranking of spectral channels for computer separation of the scene classes. Probabilities of misclassification were computed for composite signatures using four separate combinations of data source and channel selection.

Sattinger, I. J.↗
DOE OSTI2014

Modeling Pathologic Response of Esophageal Cancer to Chemoradiation Therapy Using Spatial-Temporal {sup 18}F-FDG PET Features, Clinical Parameters, and Demographics

Purpose: To construct predictive models using comprehensive tumor features for the evaluation of tumor response to neoadjuvant chemoradiation therapy (CRT) in patients with esophageal cancer. Methods and Materials: This study included 20 patients who underwent trimodality therapy (CRT + surgery) and underwent {sup 18}F-fluorodeoxyglucose (FDG) positron emission tomography/computed tomography (PET/CT) both before and after CRT. Four groups of tumor features were examined: (1) conventional PET/CT response measures (eg, standardized uptake value [SUV]{sub max}, tumor diameter); (2) clinical parameters (eg, TNM stage, histology) and demographics; (3) spatial-temporal PET features, which characterize tumor SUV intensity distribution, spatial patterns, geometry, and associated changes resulting from CRT; and (4) all features combined. An optimal feature set was identified with recursive feature selection and cross-validations. Support vector machine (SVM) and logistic regression (LR) models were constructed for prediction of pathologic tumor response to CRT, cross-validations being used to avoid model overfitting. Prediction accuracy was assessed by area under the receiver operating characteristic curve (AUC), and precision was evaluated by confidence intervals (CIs) of AUC. Results: When applied to the 4 groups of tumor features, the LR model achieved AUCs (95% CI) of 0.57 (0.10), 0.73 (0.07), 0.90 (0.06), and 0.90 (0.06). The SVM model achieved AUCs (95% CI) of 0.56 (0.07), 0.60 (0.06), 0.94 (0.02), and 1.00 (no misclassifications). With the use of spatial-temporal PET features combined with conventional PET/CT measures and clinical parameters, the SVM model achieved very high accuracy (AUC 1.00) and precision (no misclassifications)—results that were significantly better than when conventional PET/CT measures or clinical parameters and demographics alone were used. For groups with many tumor features (groups 3 and 4), the SVM model achieved significantly higher accuracy than did the LR model. Conclusions: The SVM model that used all features including spatial-temporal PET features accurately and precisely predicted pathologic tumor response to CRT in esophageal cancer.

62 RADIOLOGY AND NUCLEAR MEDICINE↗
DOE OSTI2021

Accelerated, scalable and reproducible AI-driven gravitational wave detection

The development of reusable artificial intelligence (AI) models for wider use and rigorous validation by the community promises to unlock new opportunities in multi-messenger astrophysics. Here we develop a workflow that connects the Data and Learning Hub for Science, a repository for publishing AI models, with the Hardware-Accelerated Learning (HAL) cluster, using funcX as a universal distributed computing service. Using this workflow, an ensemble of four openly available AI models can be run on HAL to process an entire month's worth (August 2017) of advanced Laser Interferometer Gravitational-Wave Observatory data in just seven minutes, identifying all four binary black hole mergers previously identified in this dataset and reporting no misclassifications. This approach combines advances in AI, distributed computing and scientific data infrastructure to open new pathways to conduct reproducible, accelerated, data-driven discovery. By combining a repository for artificial intelligence models and a supercomputing cluster, an entire month's worth of advanced LIGO data is analysed in just 7 min, finding all binary black hole mergers previously identified in this dataset and reporting no misclassifications.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2016

Submixture model to predict nepheline precipitation in waste glasses

Abstract High‐alumina high‐level waste ( HLW ) glasses are prone to nepheline precipitation during canister‐centerline cooling ( CCC ). If sufficient nepheline forms, the chemical durability of the glass will be significantly impacted. Overly conservative constraints have been developed and used to avoid the deleterious effects of nepheline formation in U.S. HLW glasses. The constraints used have been shown to significantly limit the loading of waste in glass at Hanford and therefore the cost and schedule of cleanup. A 90‐glass study was performed to develop an improved understanding of the impacts of glass composition on the formation of nepheline during CCC . The CCC crystallinity data from these glasses were combined with 657 glasses found in the literature. The trends showed significant effects of Na 2 O, Al 2 O 3 , SiO 2 , B 2 O 3 , CaO, Li 2 O, and potentially K 2 O on the propensity for nepheline formation. A pseudo‐ternary submixture model was proposed to identify the glass composition region prone to nepheline precipitation. This pseudo‐ternary with axes of SiO 2 + 1.98B 2 O 3 , Na 2 O + 0.653Li 2 O + 0.158CaO, and Al 2 O 3 was found to divide glasses that precipitate nepheline during CCC from those that do not. Application of this constraint is anticipated to increase the loading of Hanford high‐alumina HLW s in glass by roughly one‐third.

Vienna, John D.↗
DOE OSTI2014

TU-C-12A-09: Modeling Pathologic Response of Locally Advanced Esophageal Cancer to Chemo-Radiotherapy Using Quantitative PET/CT Features, Clinical Parameters and Demographics

Purpose: To develop predictive models using quantitative PET/CT features for the evaluation of tumor response to neoadjuvant chemo-radiotherapy (CRT) in patients with locally advanced esophageal cancer. Methods: This study included 20 patients who underwent tri-modality therapy (CRT + surgery) and had {sup 18}F-FDG PET/CT scans before initiation of CRT and 4-6 weeks after completion of CRT but prior to surgery. Four groups of tumor features were examined: (1) conventional PET/CT response measures (SUVmax, tumor diameter, etc.); (2) clinical parameters (TNM stage, histology, etc.) and demographics; (3) spatial-temporal PET features, which characterize tumor SUV intensity distribution, spatial patterns, geometry, and associated changes resulting from CRT; and (4) all features combined. An optimal feature set was identified with recursive feature selection and cross-validations. Support vector machine (SVM) and logistic regression (LR) models were constructed for prediction of pathologic tumor response to CRT, using cross-validations to avoid model over-fitting. Prediction accuracy was assessed via area under the receiver operating characteristic curve (AUC), and precision was evaluated via confidence intervals (CIs) of AUC. Results: When applied to the 4 groups of tumor features, the LR model achieved AUCs (95% CI) of 0.57 (0.10), 0.73 (0.07), 0.90 (0.06), and 0.90 (0.06). The SVM model achieved AUCs (95% CI) of 0.56 (0.07), 0.60 (0.06), 0.94 (0.02), and 1.00 (no misclassifications). Using spatial-temporal PET features combined with conventional PET/CT measures and clinical parameters, the SVM model achieved very high accuracy (AUC 1.00) and precision (no misclassifications), significantly better than using conventional PET/CT measures or clinical parameters and demographics alone. For groups with a large number of tumor features (groups 3 and 4), the SVM model achieved significantly higher accuracy than the LR model. Conclusion: The SVM model using all features including quantitative PET/CT features accurately and precisely predicted pathologic tumor response to CRT in esophageal cancer. This work was supported in part by National Cancer Institute Grant R21 CA131979 and R01 CA172638. Shan Tan was supported in part by the National Natural Science Foundation of China 60971112 and 61375018, and by Fundamental Research Funds for the Central Universities 2012QN086.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2022

Inference-Optimized AI and High Performance Computing for Gravitational Wave Detection at Scale

We introduce an ensemble of artificial intelligence models for gravitational wave detection that we trained in the Summit supercomputer using 32 nodes, equivalent to 192 NVIDIA V100 GPUs, within 2 h. Once fully trained, we optimized these models for accelerated inference using NVIDIA TensorRT. We deployed our inference-optimized AI ensemble in the ThetaGPU supercomputer at Argonne Leadership Computer Facility to conduct distributed inference. Using the entire ThetaGPU supercomputer, consisting of 20 nodes each of which has 8 NVIDIA A100 Tensor Core GPUs and 2 AMD Rome CPUs, our NVIDIA TensorRT-optimized AI ensemble processed an entire month of advanced LIGO data (including Hanford and Livingston data streams) within 50 s. Our inference-optimized AI ensemble retains the same sensitivity of traditional AI models, namely, it identifies all known binary black hole mergers previously identified in this advanced LIGO dataset and reports no misclassifications, while also providing a 3X inference speedup compared to traditional artificial intelligence models. We used time slides to quantify the performance of our AI ensemble to process up to 5 years worth of advanced LIGO data. In this synthetically enhanced dataset, our AI ensemble reports an average of one misclassification for every month of searched advanced LIGO data. We also present the receiver operating characteristic curve of our AI ensemble using this 5 year long advanced LIGO dataset. This approach provides the required tools to conduct accelerated, AI-driven gravitational wave detection at scale.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2017

PREPARING FOR ADVANCED LIGO: A STAR–GALAXY SEPARATION CATALOG FOR THE PALOMAR TRANSIENT FACTORY

The search for fast optical transients, such as the expected electromagnetic counterparts to binary neutron star mergers, is riddled with false positives (FPs) ranging from asteroids to stellar flares. While moving objects are readily rejected via image pairs separated by ∼1 hr, stellar flares represent a challenging foreground, significantly outnumbering rapidly evolving explosions. Identifying stellar sources close to and fainter than the transient detection limit can eliminate these FPs. Here, we present a method to reliably identify stars in deep co-adds of Palomar Transient Factory (PTF) imaging. Our machine-learning methodology utilizes the random forest (RF) algorithm, which is trained using >3×10{sup 6} sources with Sloan Digital Sky Survey (SDSS) spectra. When evaluated on an independent test set, the PTF RF model outperforms the SExtractor star classifier by ∼4%. For faint sources (r{sup ′}⩾21 mag), which dominate the field population, the PTF RF model produces a ∼19% improvement over SExtractor. To avoid false negatives in the PTF transient-candidate stream, we adopt a conservative stellar classification threshold, corresponding to a galaxy misclassification rate of 0.005. Ultimately, ∼1.70×10{sup 8} objects are included in our PTF point-source catalog, of which only ∼10{sup 6} are expected to be galaxies. We demonstrate that the PTF RF catalog reveals transients that otherwise would have been missed. To leverage its superior image quality, we additionally create an SDSS point-source catalog, which is also tuned to have a galaxy misclassification rate of 0.005. These catalogs have been incorporated into the PTF real-time pipelines to automatically reject stellar sources as non-extragalactic transients.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2018

(GO) 2 -SIM: a GCM-oriented ground-observation forward-simulator framework for objective evaluation of cloud and precipitation phase

General circulation model (GCM) evaluation using ground-based observations is complicated by inconsistencies in hydrometeor and phase definitions. Here we describe (GO) 2 -SIM, a forward simulator designed for objective hydrometeor-phase evaluation, and assess its performance over the North Slope of Alaska using a 1-year GCM simulation. For uncertainty assessment, 18 empirical relationships are used to convert model grid-average hydrometeor (liquid and ice, cloud, and precipitation) water contents to zenith polarimetric micropulse lidar and Ka-band Doppler radar measurements, producing an ensemble of 576 forward-simulation realizations. Sensor limitations are represented in forward space to objectively remove from consideration model grid cells with undetectable hydrometeor mixing ratios, some of which may correspond to numerical noise. Phase classification in forward space is complicated by the inability of sensors to measure ice and liquid signals distinctly. However, signatures exist in lidar–radar space such that thresholds on observables can be objectively estimated and related to hydrometeor phase. The proposed phase-classification technique leads to misclassification in fewer than 8 % of hydrometeor-containing grid cells. Such misclassifications arise because, while the radar is capable of detecting mixed-phase conditions, it can mistake water- for ice-dominated layers. However, applying the same classification algorithm to forward-simulated and observed fields should generate hydrometeor-phase statistics with similar uncertainty. Alternatively, choosing to disregard how sensors define hydrometeor phase leads to frequency of occurrence discrepancies of up to 40 %. So, while hydrometeor-phase maps determined in forward space are very different from model reality they capture the information sensors can provide and thereby enable objective model evaluation.

54 ENVIRONMENTAL SCIENCES↗
NASA NTRS1973

Sufficient Statistics: an Example

The feature selection problem is considered resulting from the transformation x = Bz where B is a k by n matrix of rank k and k is or = to n. Such a transformation can be considered to reduce the dimension of each observation vector z, and in general, such a transformation results in a loss of information. In terms of the divergence, this information loss is expressed by the fact that the average divergence D sub B computed using variable x is less than or equal to the average divergence D computed using variable z. If D sub B = D, then B is said to be a sufficient statistic for the average divergence D. If B is a sufficient statistic for the average divergence, then it can be shown that the probability of misclassification computed using variable x (of dimension k is or = to n) is equal to the probability of misclassification computed using variable z. Also included is what is believed to be a new proof of the well known fact that D is or = to D sub B. Using the techniques necessary to prove the above fact, it is shown that the Brattacharyya distance as measured by variable x is less than or equal to the Brattacharyya distance as measured by variable z.

Quirein, J.↗
NASA NTRS1974

On differentiating the probability of error in multipopular feature selection

A method of linear feature selection for n dimensional observation vectors which belong to one of m populations is presented. Each population has a known apriori probability and is described by a known multivariate normal density function. Specifically we consider the problem of finding a k x n matrix B of rank k (k n) for which the transformed probability of misclassification is minimized. Providing that the transformed a posterior probabilities are distinct theoretical results are obtained which, for the case k = l, give rise to a numerically tractable formula for the derivative of the probability of misclassification. It is shown that for the two population problem this condition is also necessary. The dependence of the minimum probability of error on the a priori probabilities is investigated. The minimum probability of error satisfies a uniform Lipschitz condition with respect to the a priori probabilities.

Peters, B. C.↗
NASA NTRS1974

Use of a diffuser for producing texture masks for image processing

The use of a diffuser in synthesizing binary texture-variance masks for applications in image processing is investigated. The binary texture-variance masks are used to separate various objects in an image on the basis of their spatial texture variance. The diffuser is required to reduce the speckling present in the mask. The reduction in speckling is described quantitatively as a reduction of the misclassification probability associated with the separation of the objects into several texture classes. General results are furnished for the statistics of the field scattered by the diffuser, and these are used to obtain expressions for the misclassification probabilities in the presence of a diffuser.

Stark, H.↗
NASA NTRS1975

Study of recreational land and open space using Skylab imagery

The author has identified the following significant results. A preliminary analysis of the Gratiot-Saginaw game area was conducted for the matrix probabilities of misclassification for both multispectral scanner signatures and the ERTS signatures. This analysis indicated that certain signatures of similar vegetation types had relatively high probabilities of misclassification and should logically be combined into a single signature for each major class. This would result in a single class of trees having a broad range of crown densities and two or three types of wetlands. It was clear that substantially different terrain classes could be reliably discriminated. It was also noted that areas of pine and regenerated aspen could be discriminated from other surface types.

Sattinger, I. J.↗
NASA NTRS1975

An algorithm for optimal single linear feature extraction from several Gaussian pattern classes

A computational algorithm is presented for the extraction of an optimal single linear feature from several Gaussian pattern classes. The algorithm minimizes the increase in the probability of misclassification in the transformed (feature) space. Numerical results on the application of this procedure to the remotely sensed data from the Purdue Cl flight line as well as LANDSAT data are presented. It was found that classification using the optimal single linear feature yielded a value for the probability of misclassification on the order of 30% less than that obtained by using the best single untransformed feature. Also, the optimal single linear feature gave performance results comparable to those obtained by using the two features which maximized the average divergence.

Starks, S. A.↗
NASA NTRS2001

Relative effectiveness of kinetic analysis vs single point readings for classifying environmental samples based on community-level physiological profiles (CLPP)

The relative effectiveness of average-well-color-development-normalized single-point absorbance readings (AWCD) vs the kinetic parameters mu(m), lambda, A, and integral (AREA) of the modified Gompertz equation fit to the color development curve resulting from reduction of a redox sensitive dye from microbial respiration of 95 separate sole carbon sources in microplate wells was compared for a dilution series of rhizosphere samples from hydroponically grown wheat and potato ranging in inoculum densities of 1 x 10(4)-4 x 10(6) cells ml-1. Patterns generated with each parameter were analyzed using principal component analysis (PCA) and discriminant function analysis (DFA) to test relative resolving power. Samples of equivalent cell density (undiluted samples) were correctly classified by rhizosphere type for all parameters based on DFA analysis of the first five PC scores. Analysis of undiluted and 1:4 diluted samples resulted in misclassification of at least two of the wheat samples for all parameters except the AWCD normalized (0.50 abs. units) data, and analysis of undiluted, 1:4, and 1:16 diluted samples resulted in misclassification for all parameter types. Ordination of samples along the first principal component (PC) was correlated to inoculum density in analyses performed on all of the kinetic parameters, but no such influence was seen for AWCD-derived results. The carbon sources responsible for classification differed among the variable types with the exception of AREA and A, which were strongly correlated. These results indicate that the use of kinetic parameters for pattern analysis in CLPP may provide some additional information, but only if the influence of inoculum density is carefully considered. c2001 Elsevier Science Ltd. All rights reserved.

NASA Center KSC↗