Search NASA⌕ Search

SEARCH · Search NASA

Results for “Open Set Recognition”

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.

29 records · Page 2

2021 Smoky Mountains Conference Data Challenge Synthetic-to-Real Domain Adaptation for Autonomous Driving Dataset

The dataset is comprised of both real and synthetic images from a vehicle's forward-facing camera. Each camera image is accompanied by a corresponding pixel-level semantic segmentation image (all files are .png files). In total, the dataset contains 5600 images in the training/validation set and 1400 images in the testing set. The training dataset contains mostly synthetic RGB images collected with a wide range of weather and lighting conditions using the CARLA simulator [1]. In addition, the training data also includes a small pre-selected subset of data from the Cityscapes training dataset – which is comprised of RGB-segmentation image pairs from driving scenarios in various European cities [2]. The testing data is split into three sets. The first set contains synthetic CARLA images with weather/lighting conditions that were not present in the training set. The second set is a subset of the Cityscapes testing dataset. Finally, the third set is an unknown testing set which will not be revealed to the participants until after the submission deadline. [1] Dosovitskiy, A., Ros, G., Codevilla, F., Lopez, A., and Koltun, V. (2017, October). CARLA: An open urban driving simulator. In Conference on robot learning (pp. 1-16). PMLR. [2] Cordts, M., Omran, M., Ramos, S., Rehfeld, T., Enzweiler, M., Benenson, R., ... and Schiele, B. (2016). The cityscapes dataset for semantic urban scene understanding. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 3213-3223).

99 GENERAL AND MISCELLANEOUS↗

Rapid Adaptation of Chemical Named Entity Recognition Using Few-Shot Learning and LLM Distillation

Named entity recognition (NER) has been widely used in chemical text mining for the automatic identification and extraction of chemical entities. However, existing chemical NER systems primarily focus on scenarios with abundant training data, requiring significant human effort on annotations. This poses challenges for applications in the chemical field, such as catalysis, where many advancements have traditionally relied on trial-and-error investigations and incremental adjustment of variables. This hinders catalysis science and technology progress in addressing emerging energy and environmental crises. In this work, we propose a few-shot NER model that can quickly adapt to extract new types of chemical entities by using only a limited number of annotated examples. Our model employs a metric-learning approach to transfer entity similarity knowledge from high-resource chemical domains (with abundant annotations) to enable effective entity recognition in low-resource specialized domains (limited annotation). We validate the effectiveness of our model on a few-shot chemical NER benchmark built based on six existing chemical NER data sets. Experiments show that the proposed few-shot NER model can achieve reasonable performance with only 5 examples per entity type and shows consistent improvement as the number of examples increases. Furthermore, we demonstrate how the proposed model can be trained with large language model (LLM) annotated data, opening a new pathway for rapid adaptation of NER systems. Furthermore, our approach leverages the knowledge broadness of large language models for chemistry while distilling this knowledge into a lightweight model suitable for efficient and in-house use.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Standoff Human Identification Using Body Shape

The ability to identify individuals is a key component of maintaining safety and security in public spaces and around critical infrastructure. Monitoring an open space is challenging because individuals must be identified and re-identified from a standoff distance nonintrusively, making methods like fingerprinting and even facial recognition impractical. We propose using body shape features as a means for identification from standoff sensing, either complementing other identifiers or as an alternative. An important challenge in monitoring open spaces is reconstructing identifying features when only a partial observation is available, because of the view-angle limitations and occlusion or subject pose changes. To address this challenge, we investigated the minimum number of features required for a high probability of correct identification, and we developed models for predicting a key body feature—height—from a limited set of observed features. We found that any set of nine randomly selected body measurements was sufficient to correctly identify an individual in a dataset of 4426 subjects. For predicting height, anthropometric measures were investigated for correlation with height. Their correlation coefficients and associated linear models were reported. These results—a sufficient number of features for identification and height prediction from a single feature—contribute to developing systems for standoff identification when views of a subject are limited.

Matzner, Shari↗

Computationally Efficient Multiscale Neural Networks Applied to Fluid Flow in Complex 3D Porous Media

Abstract The permeability of complex porous materials is of interest to many engineering disciplines. This quantity can be obtained via direct flow simulation, which provides the most accurate results, but is very computationally expensive. In particular, the simulation convergence time scales poorly as the simulation domains become less porous or more heterogeneous. Semi-analytical models that rely on averaged structural properties (i.e., porosity and tortuosity) have been proposed, but these features only partly summarize the domain, resulting in limited applicability. On the other hand, data-driven machine learning approaches have shown great promise for building more general models by virtue of accounting for the spatial arrangement of the domains’ solid boundaries. However, prior approaches building on the convolutional neural network (ConvNet) literature concerning 2D image recognition problems do not scale well to the large 3D domains required to obtain a representative elementary volume (REV). As such, most prior work focused on homogeneous samples, where a small REV entails that the global nature of fluid flow could be mostly neglected, and accordingly, the memory bottleneck of addressing 3D domains with ConvNets was side-stepped. Therefore, important geometries such as fractures and vuggy domains could not be modeled properly. In this work, we address this limitation with a general multiscale deep learning model that is able to learn from porous media simulation data. By using a coupled set of neural networks that view the domain on different scales, we enable the evaluation of large ( $$>512^3$$ > 512 3 ) images in approximately one second on a single graphics processing unit. This model architecture opens up the possibility of modeling domain sizes that would not be feasible using traditional direct simulation tools on a desktop computer. We validate our method with a laminar fluid flow case using vuggy samples and fractures. As a result of viewing the entire domain at once, our model is able to perform accurate prediction on domains exhibiting a large degree of heterogeneity. We expect the methodology to be applicable to many other transport problems where complex geometries play a central role.

36 MATERIALS SCIENCE↗

Three-dimensional track reconstruction for directional Dark Matter detection

Directional detection of Dark Matter is a promising search strategy. However, to perform such detection, a given set of parameters has to be retrieved from the recoiling tracks: direction, sense and position in the detector volume. In order to optimize the track reconstruction and to fully exploit the data of forthcoming directional detectors, we present a likelihood method dedicated to 3D track reconstruction. This new analysis method is applied to the MIMAC detector. It requires a full simulation of track measurements in order to compare real tracks to simulated ones. We conclude that a good spatial resolution can be achieved, i.e. sub-mm in the anode plane and cm along the drift axis. This opens the possibility to perform a fiducialization of directional detectors. The angular resolution is shown to range between 20° to 80°, depending on the recoil energy, which is however enough to achieve a high significance discovery of Dark Matter. On the contrary, we show that sense recognition capability of directional detectors depends strongly on the recoil energy and the drift distance, with small efficiency values (50%–70%). We suggest not to consider this information either for exclusion or discovery of Dark Matter for recoils below 100 keV and then to focus on axial directional data.

79 ASTRONOMY AND ASTROPHYSICS↗

A paradigm shift - the key to optimal, defensible and transparent waste management, disposal and remediation decisions - 15236

Current methods for structuring nuclear waste management, disposal, and remediation decisions are subject to three major limitations. First, they are based on a series of extraordinarily conservative assumptions. The dominant way of thinking is that, for example, conservative models yield better, safer decisions. Second, they lack methods to meaningfully address and incorporate the 'human dimension'. The dominant way of thinking is that the current approach is sufficient to bring together the values and interests of both internal and external stakeholders in making a decision. And third, they lack a formal method through which both internal and external stakeholders are encouraged collectively to 'think through' complex problems that contain multiple competing objectives so that wise, socially acceptable, and scientifically-defensible decisions are made. To overcome these limitations, a paradigm shift is needed towards a holistic, integrated approach to structure nuclear waste management, disposal and remediation decisions. The first requirement of this paradigm shift is the realization and recognition that the problems faced are decision problems, and that there is a formal theory of decision analysis that can, and should, be used to support finding the best solutions. Decision analysis, very simply, is formalized common sense. The reason it is needed for complex problems is that there are too many moving parts (variables) for humans to manage with out the formalism. Decision analysis leads to addressing problem solving using a combination of a science-based model and a costs/values model. Expectation ('realism') and uncertainty are addressed in the science-based models, whereas stakeholder values, preferences, and desires are addressed when modeling costs and value judgments. Science-based modeling tools have been developed, improved, and used since the current regulatory setting was introduced about 30 years ago. However, modeling of costs and value judgments has not often been addressed using any formal approach, despite the openings that exist in regulations (e.g., CERCLA's modifying criteria and the concept of 'as low as reasonably achievable' referenced in DOE Orders). A discussion of the need and rationale for this paradigm shift is presented, followed by a description of the technical basics of this approach to decision modeling, and ending with an explanation of the steps that are taken in this approach to structured decision-making. A software framework program called Guided Interactive Statistical Decision Tools (GiSdT - pronounced 'gist') is introduced that supports this paradigm shift. (authors)

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

PE5–PPE4–EspG 3 heterotrimer structure from mycobacterial ESX-3 secretion system gives insight into cognate substrate recognition by ESX systems

Mycobacterium tuberculosis has evolved numerous type VII secretion (ESX) systems to secrete multiple factors important for both growth and virulence across their cell envelope. ESX-1, ESX-3, and ESX-5 systems have been shown to each secrete a distinct set of substrates, including PE and PPE families of proteins, named for conserved Pro-Glu and Pro-Pro-Glu motifs in their N termini. Proper secretion of the PE–PPE proteins requires the presence of EspG, with each system encoding its own unique copy. There is no cross-talk between any of the ESX systems, and how each EspG recognizes its subset of PE–PPE proteins is currently unknown. The only current structural characterization of PE–PPE–EspG heterotrimers is from the ESX-5 system. Here we present the crystal structure of the PE5 mt –PPE4 mt –EspG 3mm heterotrimer from the ESX-3 system. Our heterotrimer reveals that EspG 3mm interacts exclusively with PPE4 mt in a similar manner to EspG 5 , shielding the hydrophobic tip of PPE4 mt from solvent. The C-terminal helical domain of EspG 3mm is dynamic, alternating between “open” and “closed” forms, and this movement is likely functionally relevant in the unloading of PE–PPE heterodimers at the secretion machinery. In contrast to the previously solved ESX-5 heterotrimers, the PE–PPE heterodimer of our ESX-3 heterotrimer is interacting with its chaperone at a drastically different angle and presents different faces of the PPE protein to the chaperone. We conclude that the PPE–EspG interface from each ESX system has a unique shape complementarity that allows each EspG to discriminate among noncognate PE–PPE pairs.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

GlycReSoft: A Software Package for Automated Recognition of Glycans from LC/MS Data

Glycosylation modifies the physicochemical properties and protein binding functions of glycoconjugates. These modifications are biosynthesized in the endoplasmic reticulum and Golgi apparatus by a series of enzymatic transformations that are under complex control. As a result, mature glycans on a given site are heterogeneous mixtures of glycoforms. This gives rise to a spectrum of adhesive properties that strongly influences interactions with binding partners and resultant biological effects. In order to understand the roles glycosylation plays in normal and disease processes, efficient structural analysis tools are necessary. In the field of glycomics, liquid chromatography/mass spectrometry (LC/MS) is used to profile the glycans present in a given sample. This technology enables comparison of glycan compositions and abundances among different biological samples, i.e. normal versus disease, normal versus mutant, etc. Manual analysis of the glycan profiling LC/MS data is extremely time-consuming and efficient software tools are needed to eliminate this bottleneck. In this work, we have developed a tool to computationally model LC/MS data to enable efficient profiling of glycans. Using LC/MS data deconvoluted by Decon2LS/DeconTools, we built a list of unique neutral masses corresponding to candidate glycan compositions summarized over their various charge states, adducts and range of elution times. Our work aims to provide confident identification of true compounds in complex data sets that are not amenable to manual interpretation. This capability is an essential part of glycomics work flows. We demonstrate this tool, GlycReSoft, using an LC/MS dataset on tissue derived heparan sulfate oligosaccharides. The software, code and a test data set are publically archived under an open source license.

Environmental Molecular Sciences Laboratory↗

A Software Defined Storage Approach to Exascale Storage Services (Final Technical Report)

The Mochi project began in 2015 in recognition of a set of changes that were set to alter the landscape of storage and I/O for DOE science. On the one hand, an influx of new classes of applications including an initial set of data intensive and machine learning applications were operating with new types of data and in new ways, as compared to traditional computational science simulation codes. Simultaneously, new memory, storage, and networking technologies were driving a need to adapt software designs to a rapidly changing design space. The goal of the three-year Mochi project was to develop and demonstrate an approach for composing storage services which provides new levels of functionality, performance, and reliability for science applications at extreme scale. The team, including a mix of Laboratory, university, and industry collaborators, focused on systems and applications of interest to DOE Office of Science, tracking developments at NERSC and the LCFs as well as application trends within ECP and SciDAC. Technical work focused on three themes: composition (understand the building blocks needed to provide storage services that meet DOE mission needs and how these components are flexibly composed), application interfacing (what service compositions are most effective for common DOE application use cases), and consistency and synchronization (taking advantage of knowledge about, and from, the system and scientific applications to provide better service than would otherwise be possible). The project was highly successful, producing 13 publications and tens of thousands of lines of open source software that is now being used by multiple ECP teams, HPC software vendors, and international collaborators. The uptake of Mochi technologies speaks to both the value of the service composition model and the quality of architecture and software developed by the team. Through continued funding under ASCR and ECP, the Mochi team and its technologies are on track to further contribute to DOE science successes into the exascale timeframe.

97 MATHEMATICS AND COMPUTING↗

COMPLEX NETWORKS IN CLIMATE SCIENCE: PROGRESS, OPPORTUNITIES AND CHALLENGES

Networks have been used to describe and model a wide range of complex systems, both natural as well as man-made. One particularly interesting application in the earth sciences is the use of complex networks to represent and study the global climate system. In this paper, we motivate this general approach, explain the basic methodology, report on the state of the art (including our contributions), and outline open questions and opportunities for future research. Datasets and systems that can be represented as interaction networks (or graphs), broadly defined as any collection of interrelated objects or entities, have received considerable attention both from a theoretical viewpoint as well as various application domains; examples include the analysis of social networks, chemical interactions between proteins, the behavior of financial markets, and many others. Recently, the study of complex networks - that is, networks which exhibit non-trivial topological properties - has permeated numerous fields and disciplines spanning the physical, social, and computational sciences. So why do networks enjoy such broad appeal? Briefly, it is their ability to serve at once as a data representation, as an analysis framework, and as a visualization tool. The analytic capabilities in particular are quite powerful, as networks can uncover structure and patterns at multiple scales, ranging from local properties to global phenomena, and thus help better understand the characteristics of complex systems. We focus on one particular application of networks in the earth sciences, namely, the construction and analysis of climate networks. Identifying and analyzing patterns in global climate is an important task of growing scientific, social, and political interest, with the goal of deepening our understanding of the complex processes underlying observed phenomena. To this end, we make the case that complex networks offer a compelling perspective for capturing the dynamics of the climate system. Moreover, the computational sciences - specifically data mining and machine learning - are able to contribute a valuable set of methods and tools ranging from pattern recognition to predictive models. Thus, in this paper we expand upon the general approach to climate networks and motivate a promising area of interdisciplinary research. Indeed, we believe that this marriage of analytic methods, computational tools and domain science has the long-term potential for a transformative impact on our understanding of the earth's climate system.

54 ENVIRONMENTAL SCIENCES↗

Muon track reconstruction in a segmented bolometric array using multi-objective optimization

Recent advances in segmented solid-state detector arrays for rare-event searches have allowed the technology to approach the ton-scale in detector mass and the scale of meters in size. Often focused around searches for neutrinoless double-beta decay or direct dark matter detection, such experiments also have the capability to search for exotic particles that leave track-like signatures across their volume. However, the segmented nature of such detector arrays often sets the spatial resolution and makes the problem of reconstructing track-like paths non-trivial. Here, in this paper, we present an algorithm that improves reconstruction of track-like events in segmented detectors using multi-objective optimization — a computational technique that optimizes more than one cost function at a time without specifying a quantitative weighting between them. Such a technique allows the reconstruction of tracks through a detector and the determination of path-lengths through individual elements. When combined with the reconstructed energy depositions in each element this allows for a calculation of the stopping power of track-like particles and opens the door to searches for particles with abnormal stopping power like monopoles or lightly-ionizing particles (LIPs). Results are presented which evaluate the precision of the reconstruction tools as they currently stand against Monte Carlo generated data. The algorithm is presented in the context of the CUORE experiment, but has applications to other segmented calorimeter detectors.

47 OTHER INSTRUMENTATION↗