Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “false negative analysis”

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
DOE OSTI2025

Identifying preferential flow from soil moisture time series: Review of methodologies

Abstract Identifying and quantifying preferential flow (PF) through soil—the rapid movement of water through spatially distinct pathways in the subsurface—is vital to understanding how the hydrologic cycle responds to climate, land cover, and anthropogenic changes. In recent decades, methods have been developed that use measured soil moisture time series to identify PF. Because they allow for continuous monitoring and are relatively easy to implement, these methods have become an important tool for recognizing when, where, and under what conditions PF occurs. The methods seek to identify a pattern or quantification that indicates the occurrence of PF. Most commonly, the chosen signature is either (1) a nonsequential response to infiltrated water, in which soil moisture responses do not occur in order of shallowest to deepest, or (2) a velocity criterion, in which newly infiltrated water is detected at depth earlier than is possible by nonpreferential flow processes. Alternative signatures have also been developed that have certain advantages but are less commonly utilized. Choosing among these possible signatures requires attention to their pertinent characteristics, including susceptibility to errors, possible bias toward false negatives or false positives, reliance on subjective judgments, and possible requirements for additional types of data. We review 77 studies that have applied such methods to highlight important information for readers who want to identify PF from soil moisture data and to inform those who aim to develop new methods or improve existing ones. Core Ideas Soil moisture data can be used to identify the occurrence of preferential flow (PF) and its initiating conditions. Various data‐analysis methods to identify PF differ in susceptibility to error, bias, and subjectivity. These methods can utilize vast amounts of data from soil moisture monitoring networks to develop understanding of when, where, and under what conditions PF occurs. Newly developed methods may lead to better accuracy and reliability, and reduce the need for subjective judgments. Plain Language Summary Preferential flow through soil occurs when a large amount of water is suddenly available, as during an intense storm. This type of flow moves rapidly through the soil in distinct narrow pathways rather than moving evenly throughout the body of soil, with major consequences for groundwater resources, ecosystems, spreading of contaminants, and other vital concerns. Methods of detecting preferential flow have been developed that utilize measurements of soil water content made by sensors installed at various depths. This measurement technology has been widely implemented, many locations now having datasets years in length, and various methods have been developed for using these to identify preferential flow. The various methods are based on different features in the soil moisture records and vary in their advantages and shortcomings. In this review, we explain and evaluate these methods, highlighting important information for their implementation to identify preferential flow from soil moisture data and for efforts to develop new methods or improve existing ones.

Nimmo, John R↗
DOE OSTI2024

Entropy removal of medical diagnostics

Shannon entropy is a core concept in machine learning and information theory, particularly in decision tree modeling. To date, no studies have extensively and quantitatively applied Shannon entropy in a systematic way to quantify the entropy of clinical situations using diagnostic variables (true and false positives and negatives, respectively). Decision tree representations of medical decision-making tools can be generated using diagnostic variables found in literature and entropy removal can be calculated for these tools. This concept of clinical entropy removal has significant potential for further use to bring forth healthcare innovation, such as quantifying the impact of clinical guidelines and value of care and applications to Emergency Medicine scenarios where diagnostic accuracy in a limited time window is paramount. This analysis was done for 623 diagnostic tools and provided unique insights into their utility. For studies that provided detailed data on medical decision-making algorithms, bootstrapped datasets were generated from source data to perform comprehensive machine learning analysis on these algorithms and their constituent steps, which revealed a novel and thorough evaluation of medical diagnostic algorithms.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Denial of Service Attack Detection via Differential Analysis of Generalized Entropy Progressions

Denial-of-Service (DoS) attacks are one the most common and consequential cyber attacks in computer networks. While existing research offers a plethora of detection methods, the issue of achieving scalability, a low false positive rate, and high detection accuracy remains open. In this work, we address this problem by developing a differential method based on generalized entropy progression. In this method, named as DoDGE, we continuously fit the line of best fit to the entropy progression of destination addresses and check if the derivative, that is, the slope of this line is less than the negative of the dynamically computed standard deviation of the derivatives. Furthermore, to distinguish from flash events, we leverage the symmetry that when a flash event occurs, the derivative of the entropy progression of source addresses is positive. With this design, we omit the usage of the thresholds and the results with five real-world network traffic datasets confirm that DoDGE outperforms threshold-based DoS attack detection by two orders of magnitude in terms of false positives on average. When compared to ten machine learning (ML) models, DoDGE achieves a balanced accuracy of 99%, while the average balanced accuracy for the ML models is 52%. Moreover, the results show that DoDGE successfully differentiates between a flash event and a DoS attack. Furthermore, since the main computation cost of DoDGE is the entropy computation, which is linear in the volume of the unit-time network flow, uses integer only operations, and works on a small fraction of the total flow, it is lightweight and scalable.

Cybersecurity, wireless communication↗
DOE OSTI2021

Validation of Scale-Derived Ages in Wild Juvenile and Adult Steelhead Using Parental-Based Tagging

Abstract Accurate age information is a critical component in fisheries monitoring, management, and research. The ability to assign accurate ages to fish by using nonlethal structures is vital in calculating cohort productivity for fish species with low abundances and variable life histories. A common nonlethal method to assign ages is by interpretation of patterns in scales. Validation of ages is not easily obtained for free-ranging fishes; however, the development of genetic “tags” has expanded age validation opportunities. In this study, we used parental-based genetic tagging to validate ages determined from scales collected from juvenile and adult steelhead Oncorhynchus mykiss in two streams within the Snake River basin. Juvenile scale ages were in agreement with 93% of the known ages. Adult scale ages were in agreement with 89% of the known ages. Although overall bias in juvenile ages was very low, we saw a slight positive bias in younger fish (young of the year and age 1) and a small negative bias in older juveniles (ages 2–4). A small negative bias in ages of adults was a consequence of errors in freshwater age. The errors observed did not significantly bias age compositions because the 90% confidence intervals about the age proportions based on scales contained the known proportions; therefore, scale analysis is an acceptable method for assigning ages to Snake River steelhead. We discuss the value of validated known-age scales as a reference collection, and as an illustrative use, we constructed a circulus count guide to aid technicians in identifying missing annuli and distinguishing true from false annuli. To demonstrate the use of the circulus guidelines, we reanalyzed samples with age discrepancies, applying the circulus count data and the known age to identify mistakes.

Reinhardt, Leslie↗
DOE OSTI2024

Inferring precocial Chinook Salmon production through single‐parentage assignments

Abstract Objective Parentage analysis is a routine methodology in fisheries research, but study systems exist where it is impractical to sample both parents. The ability to reliably assign offspring to a single parent is beneficial in these situations. We applied single‐parentage assignments to a naturally spawning population of Chinook Salmon Oncorhynchus tshawytscha to quantify production of anadromous returns by unsampled precocial males. Methods We used an approach that focused on two important aspects of parentage analyses: (1) addressing the presence of family structure within the set of sampled parents and (2) controlling for false‐positive and false‐negative assignments. Result Results indicated that 30% of reproductively successful males were precocial males, which produced 20% of the returning anadromous offspring. Conclusion This study provides a framework for applying single‐parent assignments in a salmonid study system while explicitly addressing sources of assignment errors.

Steele, Craig A.↗
DOE OSTI2021

Cric searchable image database as a public platform for conventional pap smear cytology data

Amidst the current health crisis and social distancing, telemedicine has become an important part of mainstream of healthcare, and building and deploying computational tools to support screening more efficiently is an increasing medical priority. The early identification of cervical cancer precursor lesions by Pap smear test can identify candidates for subsequent treatment. However, one of the main challenges is the accuracy of the conventional method, often subject to high rates of false negative. While machine learning has been highlighted to reduce the limitations of the test, the absence of high-quality curated datasets has prevented strategies development to improve cervical cancer screening. The Center for Recognition and Inspection of Cells (CRIC) platform enables the creation of CRIC Cervix collection, currently with 400 images (1,376 × 1,020 pixels) curated from conventional Pap smears, with manual classification of 11,534 cells. This collection has the potential to advance current efforts in training and testing machine learning algorithms for the automation of tasks as part of the cytopathological analysis in the routine work of laboratories.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2024

Predictive analytics of selections of russet potatoes

We explore the application of machine learning algorithms specifically to enhance the selection process of Russet potato (Solanum tuberosum L.) clones in breeding trials by predicting their suitability for advancement. This study addresses the challenge of efficiently identifying high-yield, disease-resistant, and climate-resilient potato varieties that meet processing industry standards. Leveraging manually collected data from trials in the state of Oregon, we investigate the potential of a wide variety of state-of-the-art binary classification models. The dataset includes 1086 clones, with data on 38 attributes recorded for each clone, focusing on yield, size, appearance, and frying characteristics, with several control varieties planted consistently across four Oregon regions from 2013 to 2021. We conduct a comprehensive analysis of the dataset that includes preprocessing, feature engineering, and imputation to address missing values. We focus on several key metrics such as accuracy, F1-score, and Matthews correlation coefficient (MCC) for model evaluation. The top-performing models, namely a feedforward neural network classifier (Neural Net), a histogram-based gradient boosting classifier (HGBC), and a support vector machine classifier (SVM), demonstrate consistent and significant results. To further validate our findings, we conducted a simulation study using the aims, data-generating mechanisms, estimands, methods, and performance measures (ADEMP) framework, simulating different data-generating scenarios to assess model robustness and performance through true positive, true negative, false positive, and false negative distributions, area under the receiver operating characteristic curve (AUC-ROC) and MCC. The simulation results highlight that non-linear models like SVM and HGBC consistently show higher AUC-ROC and MCC than logistic regression, thus outperforming the traditional linear model across various distributions, and emphasizing the importance of model selection and tuning in agricultural trials. Variable selection further enhances model performance and identifies influential features in predicting trial outcomes. The findings emphasize the potential of machine learning in streamlining the selection process for potato varieties, offering benefits such as increased efficiency, substantial cost savings, and judicious resource utilization. Our study contributes insights into precision agriculture and showcases the relevance of advanced technologies for informed decision-making in breeding programs.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2020

Data Analytics Applied to Coal Fired Boilers for Detecting Leaks

Data analytics were used to detect boiler leaks from five different coal-fired boilers including both subcritical and supercritical systems. Discriminant functions were developed that detected leaks up to two weeks prior to forced plant shutdowns for repairs. The leaks were identified to occur at different sections of the boiler for each plant, including waterwalls, economizer and superheater using conventional process measurement data. Leaking conditions were detected with a high degree of confidence (≪ 1% misclassified observations) and were able to distinguish normal operations from those time periods with steam leaks even while operating the power plants in power cycling mode.Multivariable statistical analyses, including Principal Component (PCA), cluster, and Fischer Discriminant Analysis (FDA) were used to characterize the leak occurrence. Normal and operational states with steam leaks were provided in the original process datasets. These datasets were split into two different groups for training and validation purposes. The data were sorted chronologically, and every third observation was assigned to training the Discriminant Function Model (DFM) while the rest were reserved for validation. PCA was used to reduce dimensionality of the original datasets. Canonical and FDA analyses were used to investigate the relationship between process variables. The outcome of the analyses revealed that nearly 35,000 observations were classified correctly; less than 0.05% of total observations were misclassified to be leaking, i.e. both false positives and false negatives.

Indrawan, Natarianto↗
DOE OSTI2022

Integration of Pumped Heat Energy Storage with Fossil-Fired Power Plant (Final Report)

The project team of Southwest Research Institute ® (SwRI ® ), Malta Inc. (Malta), and Luminant Generation Company LLC (Luminant) completed a feasibility study for the integration of a 100-MW, 10-hour (1000-MWh) Malta Pumped Heat Energy Storage (MPHES) system with multiple full-sized fossil-fired electricity generation units (EGU) in Luminant portfolio. MPHES is a long-duration, molten-salt-energy storage technology that uses turbomachinery and heat exchangers to transfer energy to a thermal storage media when charging, and removes the heat in a similar fashion when discharging. With high round trip efficiency (60-65%) and long lifespan (30+ years), MPHES provides economic benefits to the fossil-asset owners that can be scaled to integrate with assets across their portfolio. This technology uses hardware components, workforce personnel, and skillsets similar to those used by fossil EGUs, allowing for synergy when co-locating the two technologies. Luminant has approximately 39,000 megawatts of generation across 12 states, operating in six of the seven competitive markets in the U.S. and powered by a diverse portfolio of natural gas, nuclear, coal, and solar facilities. The DeCordova plant in Granbury, Texas, a simple cycle natural gas peaker power plant, was used as the fossil-fired asset in this project. The local market in Granbury, Texas has many influences, including several nearby power plants, a Luminant-owned nuclear plant (Comanche Peak), and substantial wind energy, which causes both negative pricing at night and high market volatility. Reducing false starts of the DeCordova plant and better responding to market volatility would be an economic advantage. Luminant is currently integrating battery storage plant on site to begin addressing these challenges. Integrating long-duration storage, like MPHES, would expand this capability beyond one hour of storage and have the potential to greatly reduce the total number of gas turbine starts. The MPHES charging requirement could help offset the overnight operating costs of Comanche Peak, which cannot load follow, and the nearby Luminant-owned Wise County combined cycle plant that cycles too often. Following the assessment of Luminant’s ERCOT-based portfolio for integration compatibility with Malta’s PHES system and the project tasks of conceptual study, technoeconomic analysis, technology gap assessment, and commercialization plan, the project team effort has resulted in several key outcomes: (1) Identification of market trends in a high-wind penetration market outside a major metropolitan area; (2) Creation of a dispatching model for the MPHES system and the pairing of Li-ion battery with a gas turbine in a real time market; (3) Revenue and cost estimations for operating MPHES alongside a gas peaker plant with real dispatching considerations and comparison with variations in the Malta implementation, including doubling the storage capacity and using two discharge drivetrains; (4) Potential carbon emission reductions possible by replacing gas turbine operation with Malta PHES operation; and (5) Summary of literature-based future market predictions for Texas.

20 FOSSIL-FUELED POWER PLANTS↗
DOE OSTI2021

Analysis of Automated Fault Detection and Diagnosis Records as an Indicator of HVAC Fault Prevalence: Methodology and Preliminary Results

Faults in commercial buildings can cause energy waste and other performance problems such as reduced occupant comfort, reduced equipment longevity, and increased noise. However, it is currently unknown how commonly faults occur in different equipment types. A method has been developed to estimate the prevalence of faults in air handling units, air terminal units, and rooftop units. This method includes two types of data. The first is data from several automated fault detection and diagnostics (AFDD) software technologies. This type of data provides a large sample that represents a wide range of building types, geographical locations, and equipment types. It includes fault diagnoses from thousands of buildings around the United States, as well as anonymized metadata describing the building and equipment characteristics. The number of fault records is in the order of 107. However, despite the size and richness of the data sample, this data contains some degree of inaccuracy, i.e., false positive and false negative findings. Therefore, the study includes a second type of data, coming from manual inspection of buildings that have had the same AFDD methods applied to them (from the commercial AFDD offerings). Since the field tests are conducted in buildings with AFDD-generated fault prevalence data, they can be combined with the larger sample size to provide insight into the potential biases or lower sensitivity of the AFDD data. Once a library of fault prevalence data is built, it will be studied to provide further insight into the drivers of fault prevalence, for example, whether prevalence is correlated with building type, geographical location (which is tied to climate and to utility rates), building size, etc. This paper describes the methods developed for this study and illustrates them with preliminary data. It discusses some of the challenges of harmonizing disparate outputs from multiple AFDD vendors, application of a unifying fault taxonomy, and fault prevalence metrics.

Ebrahimi Fakhar, Amir↗
DOE OSTI2023

A study of examiner accuracy in cartridge case comparisons. Part 2: Examiner use of the AFTE range of conclusions

Here, this report describes an analysis of how examiners used the Association of Firearm and Toolmark Examiners (AFTE) Range of Conclusions in a controlled study undertaken to estimate examiner error rates in comparing cartridge cases. Results of the error rate analysis are reported in [1]; this paper focuses on a broader analysis of how the entire collection of classification categories, especially those in the Inconclusive range, were used by the participating examiners. Volunteer active examiners with AFTE membership or working in laboratories that participate in Association of Crime Laboratory Directors (ASCLD) were provided with 15 sets of three known and one questioned cartridge cases fired from a collection of 25 new Ruger SR9 handguns. Remington 9-mm Luger (manufacturer designation L9MM3) ammunition was used and comparison sets were made up of cartridge cases fired within 100 cartridges of each other for each gun. Examiners were provided with a background survey, an answer sheet allowing for the AFTE Range of Conclusions, and return shipping materials. The participating examiners were provided with both same-source and different-source comparison sets allowing the study to assess both error rates. Responses were received from 218 participating examiners. The overall rate of false-negatives was estimated as 0.367 % from comparisons known to be from the same firearm but reported as eliminations. The overall rate of false-positives was estimated as 1.01 % from comparisons known to be from different firearms but reported as identifications. In the case of true different-source examinations, it is clear that the three Inconclusive categories and the Elimination category are not used consistently by all examiners. We identify five different apparent patterns of use of the AFTE Range of Conclusions scale, and discuss possible reasons for and implications of these differences.

99 GENERAL AND MISCELLANEOUS↗
DOE OSTI2026

Red Noise–based False Alarm Thresholds for Astrophysical Periodograms via Whittle’s Approximation to the Likelihood

Astronomers who search for periodic signals using Lomb–Scargle periodograms rely on false alarm level (FAL) estimates to identify statistically significant peaks. Although FALs are often calculated from white noise models, many astronomical time series suffer from red noise. Prewhitening is a statistical technique in which a continuum model is subtracted from the log power spectrum estimate, after which the observer can proceed with a white-noise treatment. Here we present a prewhitening-based method of calculating frequency-dependent FALs. We fit power laws and autoregressive models of order 1 to each Lomb–Scargle periodogram by minimizing the Whittle approximation to the negative log-likelihood (NLL), then calculate FALs based on the best-fit model power spectrum. Our technique is a novel extension of the Whittle NLL to datasets with uneven time sampling. We demonstrate FAL calculations using observations of α Cen B, GJ 581, HD 192310, synthetic data from the radial velocity (RV) fitting challenge, and Kepler observations of a differential rotator. The Kepler data analysis shows that only true rotation signals are detected by red noise FALs, while white noise FALs suggest all spurious peaks in the low-frequency range are significant. A high-frequency sinusoid injected into α Cen B logR$'$ HK observations exceeds the 1% red noise FAL despite having only 8.9% of the power of the dominant rotation signal. In a periodogram of HD 192310 RVs, peaks associated with differential rotation and planets are detected against the 5% red noise FAL without iterative model fitting or subtraction. The software for calculating red noise–based FALs is available on GitHub.

Astrostatistics (1882)↗
DOE OSTI2023

A comparison of machine learning methods to classify radioactive elements using prompt-gamma-ray neutron activation data

The detection of illicit radiological materials is critical to establishing a robust second line of defence in nuclear security. Neutron-capture prompt-gamma activation analysis (PGAA) can be used to detect multiple radioactive materials across the entire Periodic Table. However, long detection times and a high rate of false positives pose a significant hindrance in the deployment of PGAA-based systems to identify the presence of illicit substances in nuclear forensics. In the present work, six different machine-learning algorithms were developed to classify radioactive elements based on the PGAA energy spectra. The model performance was evaluated using standard classification metrics and trend curves with an emphasis on comparing the effectiveness of algorithms that are best suited for classifying imbalanced datasets. We analyse the classification performance based on Precision, Recall, F1-score, Specificity, Confusion matrix, ROC-AUC curves, and Geometric Mean Score (GMS) measures. The tree-based algorithms (Decision Trees, Random Forest and AdaBoost) have consistently outperformed Support Vector Machine and K-Nearest Neighbours. Based on the results presented, AdaBoost is the preferred classifier to analyse data containing PGAA spectral information due to the high recall and minimal false negatives reported in the minority class.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Evaluating various composite sampling modes for detecting pathogenic SARS-CoV-2 virus in raw sewage

Inadequate sampling approaches to wastewater analyses can introduce biases, leading to inaccurate results such as false negatives and significant over- or underestimation of average daily viral concentrations, due to the sporadic nature of viral input. To address this challenge, we conducted a field trial within the University of Tennessee residence halls, employing different composite sampling modes that encompassed different time intervals (1 h, 2 h, 4 h, 6 h, and 24 h) across various time windows (morning, afternoon, evening, and late-night). Our primary objective was to identify the optimal approach for generating representative composite samples of SARS-CoV-2 from raw wastewater. Utilizing reverse transcription-quantitative polymerase chain reaction, we quantified the levels of SARS-CoV-2 RNA and pepper mild mottle virus (PMMoV) RNA in raw sewage. Our findings consistently demonstrated that PMMoV RNA, an indicator virus of human fecal contamination in water environment, exhibited higher abundance and lower variability compared to pathogenic SARS-CoV-2 RNA. Significantly, both SARS-CoV-2 and PMMoV RNA exhibited greater variability in 1 h individual composite samples throughout the entire sampling period, contrasting with the stability observed in other time-based composite samples. Through a comprehensive analysis of various composite sampling modes using the Quade Nonparametric ANCOVA test with date, PMMoV concentration and site as covariates, we concluded that employing a composite sampler during a focused 6 h morning window for pathogenic SARS-CoV-2 RNA is a pragmatic and cost-effective strategy for achieving representative composite samples within a single day in wastewater-based epidemiology applications. This method has the potential to significantly enhance the accuracy and reliability of data collected at the community level, thereby contributing to more informed public health decision-making during a pandemic.

sampling timing↗
DOE OSTI2021

SAGE Intrusion Detection System: Sensitivity Analysis Guided Explainability for Machine Learning.

This report details the results of a three-fold investigation of sensitivity analysis (SA) for machine learning (ML) explainability (MLE): (1) the mathematical assessment of the fidelity of an explanation with respect to a learned ML model, (2) quantifying the trustworthiness of a prediction, and (3) the impact of MLE on the efficiency of end-users through multiple users studies. We focused on the cybersecurity domain as the data is inherently non-intuitive. As ML is being using in an increasing number of domains, including domains where being wrong can elicit high consequences, MLE has been proposed as a means of generating trust in a learned ML models by end users. However, little analysis has been performed to determine if the explanations accurately represent the target model and they themselves should be trusted beyond subjective inspection. Current state-of-the-art MLE techniques only provide a list of important features based on heuristic measures and/or make certain assumptions about the data and the model which are not representative of the real-world data and models. Further, most are designed without considering the usefulness by an end-user in a broader context. To address these issues, we present a notion of explanation fidelity based on Shapley values from cooperative game theory. We find that all of the investigated MLE explainability methods produce explanations that are incongruent with the ML model that is being explained. This is because they make critical assumptions about feature independence and linear feature interactions for computational reasons. We also find that in deployed, explanations are rarely used due to a variety of reason including that there are several other tools which are trusted more than the explanations and there is little incentive to use the explanations. In the cases when the explanations are used, we found that there is the danger that explanations persuade the end users to wrongly accept false positives and false negatives. However, ML model developers and maintainers find the explanations more useful to help ensure that the ML model does not have obvious biases. In light of these findings, we suggest a number of future directions including developing MLE methods that directly model non-linear model interactions and including design principles that take into account the usefulness of explanations to the end user. We also augment explanations with a set of trustworthiness measures that measure geometric aspects of the data to determine if the model output should be trusted.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Timely Reporting of Heavy Hitters Using External Memory

Given an input stream S of size N, a Φ-heavy hitter is an item that occurs at least ΦN times in S. The problem of finding heavy-hitters is extensively studied in the database literature. In this work, we study a real-time heavy-hitters variant in which an element must be reported shortly after we see its T = Φ N-th occurrence (and hence it becomes a heavy hitter). We call this the Timely Event Detection (TED) Problem. The TED problem models the needs of many real-world monitoring systems, which demand accurate (i.e., no false negatives) and timely reporting of all events from large, high-speed streams with a low reporting threshold (high sensitivity). Like the classic heavy-hitters problem, solving the TED problem without false-positives requires large space (Ω (N) words). Thus in-RAM heavy-hitters algorithms typically sacrifice accuracy (i.e., allow false positives), sensitivity, or timeliness (i.e., use multiple passes). We show how to adapt heavy-hitters algorithms to external memory to solve the TED problem on large high-speed streams while guaranteeing accuracy, sensitivity, and timeliness. Our data structures are limited only by I/O-bandwidth (not latency) and support a tunable tradeoff between reporting delay and I/O overhead. With a small bounded reporting delay, our algorithms incur only a logarithmic I/O overhead. We implement and validate our data structures empirically using the Firehose streaming benchmark. Multi-threaded versions of our structures can scale to process 11M observations per second before becoming CPU bound. In comparison, a naive adaptation of the standard heavy-hitters algorithm to external memory would be limited by the storage device’s random I/O throughput, i.e., ≈100K observations per second.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Real-Time, Adaptive Radiological Anomaly Detection and Isotope Identification Using Non-Negative Matrix Factorization

Spectroscopic anomaly detection and isotope identification algorithms are integral components in nuclear nonproliferation applications such as search operations. The task is especially challenging in the case of mobile detector systems because the observed gamma-ray background changes more than for a static detector system, and a pretrained background model can easily find itself out of domain. The result is that algorithms may exceed their intended false alarm rate or sacrifice detection sensitivity to maintain the desired false alarm rate. Non-negative matrix factorization (NMF) is a powerful tool for spectral anomaly detection and identification, but, like many similar algorithms that rely on data-driven background models, in its conventional implementation, it is unable to update in real time to account for environmental changes that affect the background spectroscopic signature. Here, we have developed a novel NMF-based algorithm that periodically updates its background model to accommodate changing environmental conditions. The adaptive NMF algorithm involves fewer assumptions about its environment, making it more generalizable than existing NMF-based methods while maintaining or exceeding detection performance on simulated and real-world datasets.

Anomaly detection↗
DOE OSTI2021

Correlations Between Panoramic Imagery and Gamma-Ray Background in an Urban Area

When searching for radiological sources in an urban area, a vehicle-borne detector system will often measure complex, varying backgrounds primarily from natural gamma-ray sources. Much work has been focused on developing spectral algorithms that retain sensitivity and minimize the false-positive rate even in the presence of such spectral and temporal variability. However, information about the environment surrounding the detector system might also provide useful clues about the expected background, which if incorporated into an algorithm, could improve performance. Recent work has focused on extensive measuring and modeling of urban areas with the goal of understanding how these complex backgrounds arise. This work presents an analysis of panoramic video images and gamma-ray background data collected in Oakland, California, by the radiological multisensor analysis platform (RadMAP) vehicle. Features were extracted from the panoramic images by semantically labeling the images and then convolving the labeled regions with the detector response. A linear model was used to relate the image-derived features to gamma-ray spectral features obtained using nonnegative matrix factorization (NMF) under different regularizations. Here we find some gamma-ray background features correlate strongly with image-derived features that measure the response-adjusted solid angle subtended by sky and buildings, and we discuss the implications for the development of future, contextually aware detection algorithms.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗