SEARCH · Engineering Papers
Results for “Online Learning”
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Improving radiograph analysis throughput through transfer learning and object detection
SIGN Fracture Care International partners with surgeons in low-resource hospitals worldwide to provide access to effective orthopedic care by donating educational materials and innovatively designed surgical implants. Over two decades, SIGN’s Online Surgical Database (SOSD) has grown to contain over 500,000 medical images, with radiographs holding the majority share. One challenge in working with hospitals worldwide is that both the radiographs uploaded to the SOSD and the data entry accompanying the uploads vary in quality. To improve the accuracy of data in the SOSD, we trained a model to detect surgical implants in radiographs. We first developed a tool to automatically detect radiographs, then trained an object detection model to determine the number and placement of surgical implants visible in the radiograph. Active learning was used to generate a training set containing 2,510 radiographs with screws, nails, and plates labeled by bounding boxes. Training a model to simultaneously recognize all three classes of implants gave a low average precision (AP) for the plate class, likely due to the low number of plate instances in our training set and the large variety of surgical plates used by SIGN-partnered surgeons. Applying standard image augmentation techniques to increase the plate count in our training set did not appreciably increase the AP of plate detection. To improve plate detection, we redrew the bounding boxes to account for correlations between the screw and plate classes. Training one model to detect nails and screws and a separate model to detect plates increased the AP of plate detection by 78.8 percentage points. The AP of each class was 80.7% for screws, 93.6% for nails, and 92.6% for plates; meanwhile, the sensitivity was 92% for screws, 86% for nails, and 81% for plates. We show that object detection methods can be used to detect surgical implants in radiographs of varying quality; however, the detection ability is dependent on the type of implant, and some implants, in our case plates, must be treated differently than others. Such tools can improve the throughput of radiograph analysis, assisting physicians and surgeons with the treatment of bone fractures.
A meta-learning based distribution system load forecasting model selection framework
This paper presents a meta-learning based, automatic distribution system load forecasting model selection framework. Furthermore, the framework includes the following processes: feature extraction, candidate model preparation and labeling, offline training, and online model recommendation. Using load forecasting needs and data characteristics as input features, multiple metalearners are used to rank the candidate load forecast models based on their forecasting accuracy. Then, a scoring-voting mechanism is proposed to weights recommendations from each meta-leaner and make the final recommendations. Heterogeneous load forecasting tasks with different temporal and technical requirements at different load aggregation levels are set up to train, validate, and test the performance of the proposed framework. Simulation results demonstrate that the performance of the meta-learning based approach is satisfactory in both seen and unseen forecasting tasks.
Selecting Post-Processing Schemes for Accurate Detection of Small Objects in Low-Resolution Wide-Area Aerial Imagery
In low-resolution wide-area aerial imagery, object detection algorithms are categorized as feature extraction and machine learning approaches, where the former often requires a post-processing scheme to reduce false detections and the latter demands multi-stage learning followed by post-processing. In this paper, we present an approach on how to select post-processing schemes for aerial object detection. We evaluated combinations of each of ten vehicle detection algorithms with any of seven post-processing schemes, where the best three schemes for each algorithm were determined using average F-score metric. The performance improvement is quantified using basic information retrieval metrics as well as the classification of events, activities and relationships (CLEAR) metrics. We also implemented a two-stage learning algorithm using a hundred-layer densely connected convolutional neural network for small object detection and evaluated its degree of improvement when combined with the various post-processing schemes. The highest average F-scores after post-processing are 0.902, 0.704 and 0.891 for the Tucson, Phoenix and online VEDAI datasets, respectively. The combined results prove that our enhanced three-stage post-processing scheme achieves a mean average precision (mAP) of 63.9% for feature extraction methods and 82.8% for the machine learning approach.
A machine learning approach to water quality forecasts and sensor network expansion: Case study in the Wabash River Basin, United States
Abstract Midwestern cities require forecasts of surface nitrate loads to bring additional treatment processes online or activate alternative water supplies. Concurrently, networks of nitrate monitoring stations are being deployed in river basins, co‐locating water quality observations with established stream gauges. However, tools to evaluate the future value of expanded networks to improve water quality forecasts remains challenging. Here, we construct a synthetic data set of stream discharge and nitrate for the Wabash River Basin—one of the United States’ most nutrient polluted basins—using the established Agro‐IBIS and THMB models. Synthetic data enables rapid, unbiased and low‐cost assessment of potential sensor placements to support management objectives, such as near‐term forecasting. Using the synthetic data, we established baseline 1‐day forecasts for surface water nitrate at 12 cities in the basin using support vector machine regression (SVMR; RMSE 0.48–3.3 ppm). Next, we used the SVMRs to evaluate the improvement in forecast performance associated with deployment of additional nitrate sensors. We identified the optimal sensor placement to improve forecasts at each city, and the relative value of sensors at each candidate location. Finally, we assessed the co‐benefit realized by other cities when a sensor is deployed to optimize a forecast at one city, finding significant positive externalities in all cases. Ultimately, our study explores the potential for machine learning to make near‐term predictions and critically evaluate the improvement realized by expanding a monitoring network. While we use nitrate pollution in the Wabash River Basin as a case study, this approach could be readily applied to any problem where the future value of sensors and network design are being evaluated.
Energy and Emission Prediction for Mixed-Vehicle Transit Fleets Using Multi-task and Inductive Transfer Learning
Public transit agencies are focused on making their fixed-line bus systems more energy efficient by introducing electric (EV) and hybrid (HV) vehicles to their fleets. However, because of the high upfront cost of these vehicles, most agencies are tasked with managing a mixed-fleet of internal combustion vehicles (ICEVs), EVs, and HVs. In managing mixed-fleets, agencies require accurate predictions of energy use for optimizing the assignment of vehicles to transit routes, scheduling charging, and ensuring that emission standards are met. The current state-of-the-art is to develop separate neural network models to predict energy consumption for each vehicle class. Although different vehicle classes’ energy consumption depends on a varied set of covariates, we hypothesize that there are broader generalizable patterns that govern energy consumption and emissions. In this paper, we seek to extract these patterns to aid learning to address two problems faced by transit agencies. First, in the case of a transit agency which operates many ICEVs, HVs, and EVs, we use multi-task learning (MTL) to improve accuracy of forecasting energy consumption. Second, in the case where there is a significant variation in vehicles in each category, we use inductive transfer learning (ITL) to improve predictive accuracy for vehicle class models with insufficient data. As this work is to be deployed by our partner agency, we also provide an online pipeline for joining the various sensor streams for fixed-line transit energy prediction. Here, we find that our approach outperforms vehicle-specific baselines in both the MTL and ITL settings.
Deep Learning Parameterization of Vertical Wind Velocity Variability via Constrained Adversarial Training
Atmospheric models with typical resolution in the tenths of kilometers cannot resolve the dynamics of air parcel ascent, which varies on scales ranging from tens to hundreds of meters. Small-scale wind fluctuations are thus characterized by a subgrid distribution of vertical wind velocity W with standard deviation σ W . The parameterization of σ W is fundamental to the representation of aerosol–cloud interactions, yet it is poorly constrained. Using a novel deep learning technique, this work develops a new parameterization for σ W merging data from global storm-resolving model simulations, high-frequency retrievals of W , and climate reanalysis products. The parameterization reproduces the observed statistics of σ W and leverages learned physical relations from the model simulations to guide extrapolation beyond the observed domain. Incorporating observational data during the training phase was found to be critical for its performance. The parameterization can be applied online within large-scale atmospheric models, or offline using output from weather forecasting and reanalysis products.
High performance FPGA embedded system for machine learning based tracking and trigger in sPhenix and EIC
We present a comprehensive end-to-end pipeline to classify triggers versus background events in this paper. This pipeline makes online decisions to select signal data and enables the intelligent trigger system for efficient data collection in the Data Acquisition System (DAQ) of the upcoming sPHENIX and future EIC (Electron-Ion Collider) experiments. Starting from the coordinates of pixel hits that are lightened by passing particles in the detector, the pipeline applies three-stage of event processing (hits clustering, track reconstruction, and trigger detection) and labels all processed events with the binary tag of trigger versus background events. The pipeline consists of deterministic algorithms such as clustering pixels to reduce event size, tracking reconstruction to predict candidate edges, and advanced graph neural network-based models for recognizing the entire jet pattern. In particular, we apply the message-passing graph neural network to predict links between hits and reconstruct tracks and a hierarchical pooling algorithm (DiffPool) to make the graph-level trigger detection. We obtain an impressive performance (≥70% accuracy) for trigger detection with only 3200 neuron weights in the end-to-end pipeline. We deploy the end-to-end pipeline into a field-programmable gate array (FPGA) and accelerate the three stages with speedup factors of 1152, 280, and 21, respectively.
Mission-Based Serious Games for Cross-Cultural Communication Training
Appropriate cross-cultural communication requires a critical skill set that is increasingly being integrated into regular military training regimens. By enabling a higher order of communication skills, military personnel are able to interact more effectively in situations that involve local populations, host nation forces, and multinational partners. The Virtual Cultural Awareness Trainer (VCAT) is specifically designed to help address these needs. VCAT is deployed by Joint Forces Command (JFCOM) on Joint Knowledge Online (JKO) as a means to provide online, mission-based culture and language training to deploying and deployed troops. VCAT uses a mix of game-based learning, storytelling, tutoring, and remediation to assist in developing the component skills required for successful intercultural communication in mission-based settings.
Understanding and Explicitly Measuring Linguistic and Stylistic Properties of Deception via Generation and Translation
Massive digital disinformation is one of the main risks of modern society. Hundreds of models and linguistic analyses have been done to compare and contrast misleading and credible content online. However, most models do not remove the confounding factor of a topic or narrative when training, so the resulting models learn a clear topical separation for misleading versus credible content. We study the feasibility of using two strategies to disentangle the topic bias from the models to understand and explicitly measure linguistic and stylistic properties of content from misleading versus credible content. First, we develop conditional generative models to create news content that is characteristic of different credibility levels. We perform multi-dimensional evaluation of model performance on mimicking both the style and linguistic differences that distinguish news of different credibility using machine translation metrics and classification models. We show that even though generative models are able to imitate both the style and language of the original content, additional conditioning on both the news category and the topic leads to reduced performance. In a second approach, we perform deception style ``transfer" by translating deceptive content into the style of credible content and vice versa. Extending earlier studies, we demonstrate that, when conditioned on a topic, deceptive content is shorter, less readable, more biased, and more subjective than credible content, and transferring the style from deceptive to credible content is more challenging than the opposite direction.
A transfer learning approach to energy-efficient control of small and medium-sized commercial buildings
Model-free reinforcement learning (RL) provides a data-driven and adaptive approach to optimize building energy use while satisfying occupant comfort. This powerful tool does not need any prior knowledge about the environment and system it is optimizing and can adapt its policy based on the changes in captures. Like any other data-driven tool, it faces high training costs due to the extensive agent-environment interactions required to capture long-term building dynamics and user comfort. Transfer learning, particularly policy distillation, offers a promising way to accelerate training by leveraging pretrained RL agents in different building and system types. Here, this study investigates online student distillation, in which the student model updates its neural network weights using outputs from teacher models. The work introduces a student distillation strategy designed for efficient knowledge transfer, along with a teacher selection method that ensures high-quality guidance. The approach is validated using a highly calibrated whole building energy model for a small/medium commercial building test facility. Results show substantial reductions in training time and data requirements while surpassing the performance of ASHRAE Guideline 36, an advanced rule-based control strategy. The distilled RL model required 45% less data and achieved 20% higher cumulative rewards than a state-of-the-art RL model, with faster convergence and lower energy consumption. These outcomes demonstrate that effective transfer learning enables a scalable and data-efficient energy management solution for commercial buildings.
User Role Identification in Software Vulnerability Discussions over Social Networks
Understanding and early awareness of software vulnerabilities is vital for preventing and mitigating potential impacts from cybersecurity events. One step toward early characterization of software vulnerabilities may involve analyzing discussion and spread of information in online social networks. Prior work has used information from such discussions over multiple online forums to develop dynamic networks among users followed by analysis of structure, spread, and information evolution. In this work, we advance the state-of-the-art by focusing on data-driven learning of types, roles, and transition of roles exhibited by users over time. In social networks, users take on particular roles based on their actions and structure of the network. Identifying “meaningful” roles can help separate potential users of interest from the larger community, and identify patterns in a network. We will identify and compare roles found in online forums (e.g., Twitter) using techniques such as feature-based Non-negative Matrix Factorization coupled with topological and influence-based measures of centrality. Since users’ activities change over time, we also analyze role evolution in dynamic networks.
Open Chemistry, JupyterLab, REST, and quantum chemistry
Quantum chemistry must evolve if it wants to fully leverage the benefits of the internet age, where the worldwide web offers a vast tapestry of tools that enable users to communicate and interact with complex data at the speed and convenience of a button press. The Open Chemistry project has developed an open-source framework that offers an end-to-end solution for producing, sharing, and visualizing quantum chemical data interactively on the web using an array of modern tools and approaches. These tools build on some of the best open-source community projects such as Jupyter for interactive online notebooks, coupled with 3D accelerated visualization, state-of-the-art computational chemistry codes including NWChem and Psi4, and emerging machine learning and data mining tools such as ChemML and ANI. They offer flexible formats to import and export data, along with approaches to compare computational and experimental data.
The PMDP Roadmap
NASA's complex and highly technical missions rely on effective project teams and managers. Since 1993, through its Project Management Development Process (PMDP), the Academy of Program and Project Leadership (APPL) has offered direction to the Agency's project practitioners as they advance in their careers. PMDP helps identify and sequence professional experiences, courses, and other project-based learning experiences that support individual career goals and center activities by outlining competencies at four levels of development. The result is that PMDP provides NASA project practitioners with a road map to the knowledge and competencies appropriate for their job and the jobs to which they aspire. Plus, new this year, APPL has rolled out its electronic Project Management Development Process (ePMDP) tool, a learning management system that includes a dynamic presentation of the PMDP levels, competency areas, competency organizational structures, Individual Development Plans (IDP), and online PMDP enrollment. APPL's website, www.appl.nasa.gov, provides access to ePMDP as well as other online resources for NASA practitioners enrolled in the Project Management Development Process.
A Real-Time Safety and Quality Reporting System: Assessment of Clinical Data and Staff Participation
Purpose: To report on the use of an incident learning system in a radiation oncology clinic, along with a review of staff participation. Methods and Materials: On September 24, 2010, our department initiated an online real-time voluntary reporting system for safety issues, called the Radiation Oncology Quality Reporting System (ROQRS). We reviewed these reports from the program's inception through January 18, 2013 (2 years, 3 months, 25 days) to assess error reports (defined as both near-misses and incidents of inaccurate treatment). Results: During the study interval, there were 60,168 fractions of external beam radiation therapy and 955 brachytherapy procedures. There were 298 entries in the ROQRS system, among which 108 errors were reported. There were 31 patients with near-misses reported and 27 patients with incidents of inaccurate treatment reported. These incidents of inaccurate treatment occurred in 68 total treatment fractions (0.11% of treatments delivered during the study interval). None of these incidents of inaccurate treatment resulted in deviation from the prescription by 5% or more. A solution to the errors was documented in ROQRS in 65% of the cases. Errors occurred as repeated errors in 22% of the cases. A disproportionate number of the incidents of inaccurate treatment were due to improper patient setup at the linear accelerator (P<.001). Physician participation in ROQRS was nonexistent initially, but improved after an education program. Conclusions: Incident learning systems are a useful and practical means of improving safety and quality in patient care.
Hampshire College Center for Science Education. Final Report on Activities Supported by the Department of Energy Grant No. DE-FG02-06ER64256
Hampshire College's Center for Science Education (Center) focuses on teacher professional development, curriculum development, and student enrichment programs. The Center also maintains research programs on teacher change, student learning and instructional effectiveness. The Center's work promotes learning that persists over time and transfers to new situations in and out of school. The projects develop the implications of the increasing agreement among teachers and researchers that effective learning involves active concept mastery and consistent practice with inquiry and critical thinking. The Center's objective is to help strengthen the pipeline of U.S. students pursuing postsecondary study in STEM fields. The Center achieves this by fostering an educational environment in which science is taught as an active, directly experienced endeavor across the K-16 continuum. Too often, young people are dissuaded from pursuing science because they do not see its relevance, instead experiencing it as dry, rote, technical. In contrast, when science is taught as a hands-on, inquiry-driven process, students are encouraged to ask questions grounded in their own curiosity and seek experimental solutions accordingly. In this way, they quickly discover both the profound relevance of science to their daily lives and its accessibility to them. Essentially, they learn to think and act like real scientists. The Center’s approach is multi-faceted: it includes direct inquiry-based science instruction to secondary and postsecondary students, educating the next generation of teachers, and providing new educational opportunities for teachers already working in the schools. Funding from the Department of Energy focused on the last population, enabling in-service teachers to explore and experience the pedagogy of inquiry-based science for themselves, and to take it back to their classrooms and students. The Center has demonstrated that the inquiry-based approach to science learning is compatible with existing state curriculum frameworks and produces students who understand and are positively inclined toward science. Funds from this Department of Energy grant supported three projects that involved K-16 science outreach: 1. Teaching Issues and Experiments in Ecology (TIEE). TIEE a peer-reviewed online journal and curriculum resource for postsecondary science teachers. 2. The Collaboration for Excellence in Science Education (CESE). CESE is a partnership with the Amherst, Massachusetts school system to foster the professional development of science teachers, and to perform research on student learning in the sciences and on teacher change. The project draws on Hampshire's long experience with inquiry-oriented and interdisciplinary education, as well as on its unique strengths in cognitive science. The project is run as design research, working with teachers to improve their practices and studying student and/or teacher outcomes. 3. Day in the Lab. Grant funds partially supported the expansion of the ongoing science outreach activities of the School of Natural Science. These activities are focused on local districts with large minority enrollments, including the Amherst, Holyoke and Springfield Public School Districts, and the Pioneer Valley Performing Arts Charter School (PVPA). Each of the three projects supported by the grant met or exceeded its goals. In part, the successes we met were due to continuity and communication among the staff of the programs. At the beginning of the CESE project, a science outreach coordinator was recruited. He worked throughout the grant period along with a senior researcher and the project's curriculum director. Additionally, the director and an undergraduate student conducted research on teacher change. The science outreach coordinator acted as a liaison among Hampshire College, the school districts, and a number of local businesses and agencies, providing organizational support, discussion facilitation, classroom support for teachers, and materials purchase. His presence in the schools kept teachers engaged and supported. He also brought the PVPA Charter School into the project. He worked closely with the educational outreach coordinator at Hampshire who oversaw the Day in the Lab program. Together, they have ensured the continuity of support to the schools through the use and placement of student interns. Finally, the director and coordinators worked with the Hitchock Center for the Environment to bring the two science professional development efforts in Amherst together. The joint development of workshops for elementary teachers was extremely successful. A major reason for the successes of the CESE program was the strength of the teacher outreach team and the sheer number of hours spent building relationships, talking about teaching and learning, planning projects, developing curriculum, and working with experts throughout the Pioneer Valley.
Two-Column Aerosol Project (TCAP) Field Campaign Report
This study included the deployment of the Department of Energy (DOE) Atmospheric Radiation Measurement (ARM) Climate Research Facility’s Mobile Facility (AMF), ARM Mobile Aerosol Observing System (MAOS) and the ARM Aerial Facility (AAF). The study was a collaborative effort involving scientists from DOE national laboratories, NOAA, NASA, and universities. The AAF and MAOS were deployed for two approximately month-long Intensive Operational Periods (IOPs) conducted in June 2012 and February 2013. Seasonal differences in the aerosol chemical and optical properties observed using the AMF, AAF, and MAOS are presented in this report. The total mass loading of aerosol is found to be much greater in the summer than in the winter, with the difference associated with greater amounts of organic aerosol. The mass fraction of organic aerosol is much reduced in the winter, when sulfate is the dominant aerosol type. Surprisingly, very little sea-salt aerosol was observed in the summer. In contrast, much more sea salt aerosol was observed in the winter. The mass loading of black carbon is nearly the same in both seasons. These differences lead to a relative increase in the aerosol light absorption in the winter and an associated decrease in observed single-scattering albedo. Measurements of aerosol mixing state were made using a single-particle mass spectrometer, which showed that the majority of the summertime aerosol consisted of organic compounds mixed with various amounts of sulfate. A number of other findings are also summarized in the report, including: impact of aerosol layers aloft on the column aerosol optical depth; documentation of the aerosol properties at the AMF; differences in the aerosol properties associated with both columns, which are not systematic but reflect the complicated meteorological and chemical processes that impact aerosol as it is advected away from North America; and new instruments and data-processing techniques for measuring both aerosol and cloud properties that were deployed for the first time during the TCAP. Key lessons learned during TCAP include the need for closer coordination between the AMF, MAOS, and the AAF so that all AMF instruments can be online and functioning during the AAF IOPs. Based on experiences from TCAP, it is also important for instrument mentors, or other relevant individuals, to review data on a regular basis to ensure that data quality remains high during the entire deployment. TCAP was marked by two important meteorological events including the passage of Hurricane Sandy at the end of October 2012 and the occurrence of one of the largest New England blizzards in recorded history. During Sandy the AMF received some, generally minor, damage and was largely functional a short time after the storm. The blizzard led to extensive power outages on Cape Cod and a multi-day interruption of measurements by the AMF, MAOS, and AAF. In each case, however, the ARM Facilities were returned to service and functioning as soon as was reasonably possible.
Speaker-targeted Synthetic Speech Detection
Text-to-speech technologies are evolving quickly towards realistic-sounding human-like voices. As this technology improves, so does the opportunity for malpractice in speaker identification (SID) via spoofing, the process of impersonating a voice biometric via synthesis. More data typically equates to a more realistic voice model, which poses an issue for well-known subjects, such as politicians and celebrities, who have vast amounts of multimedia available online. Detection of synthetic speech has relied on signal processing techniques that focus on the generation of new acoustic features and train deep learning models to detect when an audio file has been manipulated through the characterization of unnatural changes or artifacts. However, these techniques do not use any information from the speaker they are evaluating. This paper proposes to incorporate information from the speaker-of-interest (SoI) into the models to avoid specific spoofing attacks for certain vulnerable people. The wealth of data for well-known people can also be used to train a speaker-specific spoofing detector with a higher level of accuracy than a speaker-independent model. The paper proposes a new xResNet-PLDA system and compares it to three different baseline systems: a state-of-the-art speaker identification system, an xResNet system trained to discriminate between bona fide and fake speech, and a speaker identification system in which the PLDA and calibration models were trained with bona fide and fake speech. We evaluated the systems in two different scenarios — a cross-validation scenario and a hold-out scenario — with three different databases. We show how the proposed system outperforms dramatically the baseline systems in each scenario and for each database. Finally, we show how using a small amount of the SoI’s speech to adapt global calibration parameters improves the performance of the system, especially in unseen conditions.