Search NASA⌕ Search

SEARCH · Search NASA

Results for “multiple instance 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.

At least 19 records
All sourcesNASA NTRSDOE OSTI
NASA NTRSDate not supplied

Predicting Adverse Events and their Precursors in Aviation Using Multi-Class Multiple-Instance Learning

In recent years, there has been a rapid growth in the application of machine learning techniques that leverage aviation data collected from commercial airline operations to improve safety. Anomaly detection and predictive maintenance have been the main targets for machine learning applications. However, this paper focuses on the identification of precursors, which is a relatively newer application. Precursors are events correlated with adverse events that happen prior to the adverse event itself. Therefore, precursor mining provides many benefits including understanding the reasons behind a safety incident and the ability to identify signatures, which can be tracked throughout a flight to alert the operators of an potential upcoming adverse event. This work proposes using the multiple-instance learning (MIL) framework, a weakly supervised learning task, combined with a carefully designed Multi-Head Convolutional Neural Networks-Recurrent Neural Networks (MHCNN-RNN) architecture to predict different type of adverse events for any given flights and identify their precursors with little to no post-processing.Results obtained show that the MHCNN-RNN is able to accurately forecast high speed and high path angle events during the approach, and that it is also capable of determining the aircraft’s parameters that are correlated to these events. These parameters can be considered precursors to the events.

multiple instance learning↗
NASA NTRS2019

Deep Learning Method for Detecting Precursors to Adverse Events

With the recent advancements in Deep Learning methods, the ability to model large complex heterogeneous data sets are fundamentally changing industry and research. Coupled with hardware improvements, and ease of implementation, a wide variety of deep neural network architectures can quickly be developed to solve a sweeping range of problems such as: object detection in images, automatic healthcare diagnosis using heterogenous data sources, real time language translating and sentence prediction, upscaling low resolution images, and forecasting of multivariate timeseries. Generally, many of these architectures outperform classical machine learning approaches in their respective tasks, however, this typically comes at a cost of interpretability. These black box algorithms generally suffer from lack of transparency in both model complexity as well as the rationale behind the prediction. This lack of comprehension, is driving an emerging area of interest in “Explainable AI”. An algorithm called: “Deep Temporal Multiple Instance Learning”1 was a recently developed to identify precursors to adverse events and has been applied in the aviation domain. The deep learning architecture is designed to capture the evolution of the probability of the outcome over the time preceding the adverse event using a multiple instance learning approach as illustrated in Figure 1. Precursors are defined when the probability of the event has exceeded a threshold at some point in the timeseries, at which point, a sensitivity analysis is performed to determine contributing factors. The contributing factors are used to explain and define the precursor during the periods where the probability score is high. The identified contributing factors are then presented to subject matter experts to provide objective insights into the leading factors associated with the particular adverse event. The algorithm has been tested on flight data from a commercial airline and has the ability to discover precursors to known adverse events that take the form of safety critical operations, such as unstable approach events on final approach. Apart from detecting precursors to adverse events, the converse can also be leveraged to discover corrective actions. These positive actions manifest themselves as periods in the timeseries when the precursor score has been lowered from an elevated state; meaning that if the system had been left uncorrected, it would have eventually reached the adverse event state. Characterizing these state changes can help identify successful interventions that may not have been known before. Policy makers and procedure designers can use this additional knowledge to craft more safety and efficient resilient procedures for future operations and therefore improve the overall performance of the National Airspace.

Matthews, Bryan L.↗
NASA NTRS2007

Salience Assignment for Multiple-Instance Regression

We present a Multiple-Instance Learning (MIL) algorithm for determining the salience of each item in each bag with respect to the bag's real-valued label. We use an alternating-projections constrained optimization approach to simultaneously learn a regression model and estimate all salience values. We evaluate this algorithm on a significant real-world problem, crop yield modeling, and demonstrate that it provides more extensive, intuitive, and stable salience models than Primary-Instance Regression, which selects a single relevant item from each bag.

regression↗
NASA NTRS2010

Salience Assignment for Multiple-Instance Data and Its Application to Crop Yield Prediction

An algorithm was developed to generate crop yield predictions from orbital remote sensing observations, by analyzing thousands of pixels per county and the associated historical crop yield data for those counties. The algorithm determines which pixels contain which crop. Since each known yield value is associated with thousands of individual pixels, this is a multiple instance learning problem. Because individual crop growth is related to the resulting yield, this relationship has been leveraged to identify pixels that are individually related to corn, wheat, cotton, and soybean yield. Those that have the strongest relationship to a given crop s yield values are most likely to contain fields with that crop. Remote sensing time series data (a new observation every 8 days) was examined for each pixel, which contains information for that pixel s growth curve, peak greenness, and other relevant features. An alternating-projection (AP) technique was used to first estimate the "salience" of each pixel, with respect to the given target (crop yield), and then those estimates were used to build a regression model that relates input data (remote sensing observations) to the target. This is achieved by constructing an exemplar for each crop in each county that is a weighted average of all the pixels within the county; the pixels are weighted according to the salience values. The new regression model estimate then informs the next estimate of the salience values. By iterating between these two steps, the algorithm converges to a stable estimate of both the salience of each pixel and the regression model. The salience values indicate which pixels are most relevant to each crop under consideration.

Wagstaff, Kiri L.↗
DOE OSTI2025

Protein-ligand binding affinity prediction using multi-instance learning with docking structures

Recent advances in 3D structure-based deep learning approaches demonstrate improved accuracy in predicting protein-ligand binding affinity in drug discovery. These methods complement physics-based computational modeling such as molecular docking for virtual high-throughput screening. Despite recent advances and improved predictive performance, most methods in this category primarily rely on utilizing co-crystal complex structures and experimentally measured binding affinities as both input and output data for model training. Nevertheless, co-crystal complex structures are not readily available and the inaccurate predicted structures from molecular docking can degrade the accuracy of the machine learning methods. We introduce a novel structure-based inference method utilizing multiple molecular docking poses for each complex entity. Our proposed method employs multi-instance learning with an attention network to predict binding affinity from a collection of docking poses. We validate our method using multiple datasets, including PDBbind and compounds targeting the main protease of SARS-CoV-2. The results demonstrate that our method leveraging docking poses is competitive with other state-of-the-art inference models that depend on co-crystal structures. This method offers binding affinity prediction without requiring co-crystal structures, thereby increasing its applicability to protein targets lacking such data.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Multi-head physics-informed neural networks for learning functional priors and uncertainty quantification

In numerous applications, the integration of prior knowledge and historical information is essential, particularly for tasks requiring the solution of ordinary or partial differential equations (ODEs/PDEs) in data-sparse or noisy environments. For instance, achieving accurate solutions to time-dependent PDEs with limited initial condition measurements necessitates an effective strategy for embedding prior knowledge. Hard-parameter sharing architectures in neural networks (NNs) have demonstrated success in both traditional and scientific machine learning domains, facilitating the learning of informative representations. Here, in this study, we introduce a novel, yet efficient, method to enhance physics-informed neural networks (PINNs) by incorporating a multi-head structure that enables the learning of functional priors from both empirical data and governing physical laws. This prior information can then be used to address data sparsity and high-level noise in solving ODE/PDE problems with uncertainty quantification (UQ). The approach, termed Multi-Head PINN (MH-PINN), consists of a shared body NN and multiple head NNs, each corresponding to an individual PINN instance. Our framework for functional prior learning is carried out in two stages: (1) training the MH-PINNs to develop a shared body NN alongside multiple head NNs, and (2) employing these trained head NNs to estimate a prior distribution through a normalizing flow-based density estimator. The learned functional prior can then be applied as a regularization mechanism in deterministic contexts or as an informative prior within a Bayesian inference framework, aiding in the resolution of subsequent ODE/PDE tasks. We evaluate the efficacy of MH-PINNs across five benchmark problems, including a high-dimensional parametric PDE, all characterized by data sparsity or substantial noise levels. Our findings reveal that MH-PINNs deliver accurate solutions and robust UQ, demonstrating adaptability across a range of complex and challenging scenarios.

Bayesian inference↗
DOE OSTI2026

Feature-Based PMU Event Classification under Variable PMU Participation and Overlapping Events

Danovo Energy Solution's presented its paper named: Feature-Based PMU Event Classification under Variable PMU Participation and Overlapping Events at the 2026 Georgia Tech Fault & Disturbance Analysis Conference. The full paper can be found at OSTI ID# 3169150 Paper Abstract—Phasor Measurement Units (PMUs) stream time synchronized, high-resolution measurements from the grid, enabling data-driven techniques for event detection and classification. Accurate event classification improves grid reliability and stability. Events can be detected by varying numbers of PMUs and exhibit different durations depending on the event type. This variability challenges standard classifiers that require uniform input sizes. Moreover, multiple events may coincide, which increases classification complexity. Standard classifiers assign each instance to the class with the highest predicted probability, whereas overlapping events may exhibit comparable probabilities across multiple classes. In this study, to handle data size variability, we extract a wide range of time–frequency domain features from all available PMUs for each event into a fixed-length vector, facilitating the application of standard machine learning classifiers, including Random Forest, XGBoost, LightGBM, Support Vector Machine, and Multilayer Perceptron. To account for overlapping events, a probabilistic post-processing step is applied. For a given data instance, if multiple predicted class probabilities exceed 30% and the differences between them are less than 10%, the event is assigned to multiple classes. Experiments using real-world PMU data demonstrate that the Random Forest and XGBoost models achieve the highest accuracy, while the proposed post-processing method yields perfect classification performance on external unseen test sets.

Nematirad, Reza [Danova Energy Solutions]↗
DOE OSTI2026

Feature-Based PMU Event Classification under Variable PMU Participation and Overlapping Events

This paper is the basis for a presentation help at the 2026 Georgia Tech Fault & Disturbance Analysis Conference, which can be found at OSTI # 3168287 Paper Abstract—Phasor Measurement Units (PMUs) stream time synchronized, high-resolution measurements from the grid, enabling data-driven techniques for event detection and classification. Accurate event classification improves grid reliability and stability. Events can be detected by varying numbers of PMUs and exhibit different durations depending on the event type. This variability challenges standard classifiers that require uniform input sizes. Moreover, multiple events may coincide, which increases classification complexity. Standard classifiers assign each instance to the class with the highest predicted probability, whereas overlapping events may exhibit comparable probabilities across multiple classes. In this study, to handle data size variability, we extract a wide range of time–frequency domain features from all available PMUs for each event into a fixed-length vector, facilitating the application of standard machine learning classifiers, including Random Forest, XGBoost, LightGBM, Support Vector Machine, and Multilayer Perceptron. To account for overlapping events, a probabilistic post-processing step is applied. For a given data instance, if multiple predicted class probabilities exceed 30% and the differences between them are less than 10%, the event is assigned to multiple classes. Experiments using real-world PMU data demonstrate that the Random Forest and XGBoost models achieve the highest accuracy, while the proposed post-processing method yields perfect classification performance on external unseen test sets.

Nematirad, Reza [Danovo Energy Solutions]↗
DOE OSTI2026

The Hydropower Game: An Interactive Learning Tool for the Future Hydropower Workforce

This report documents a Seedling project supported by the U.S. Department of Energy (DOE) Water Power Technologies Office (WPTO), conducted by Argonne National Laboratory (ANL) and the National Laboratory of the Rockies (NLR). The Seedling and Sapling Program provides small grants over short duration to enable early-stage research at national laboratories. This funding mechanism is intended to cultivate innovative ideas and expand research road maps in hydropower and marine energy technologies. Under this Seedling award, the project team developed an educational outreach tool, or “serious game”, built in Python and powered by the Pygame library, aimed at teaching fundamental hydropower concepts through an engaging, interactive learning experience. The game features multiple lessons covering several hydropower topics while allowing for expansion and customization in the instance of future funding availability: • Hydropower Plant Types: Players explore and compare the mechanics and applications of run-of-river, dam hydropower, and pumped-storage hydropower. • Flow Dynamics and Power Generation: Interactive tasks demonstrate how release rate and hydraulic head combine to determine power output. • Grid Operations and Load Following: Simulations illustrate how hydropower facilities respond to fluctuating electricity demand to balance the grid. • Market Integration: Levels scaffold understanding of how hydropower interfaces with the broader energy market, including operational and economic considerations. The tool was showcased at several science, technology, engineering, and mathematics (STEM) outreach events, where it was demoed to students, educators, and the general public. These events provided valuable opportunities to collect feedback on the gameplay, storyline, and educational objectives from a wide audience. The game was presented at STEMapalooza, Introduce a Girl to Engineering Day, STEMCON, and Clean Currents 2025. Insights gathered during these demonstrations informed refinements to the game’s user experience and strengthened its effectiveness as an educational tool for teaching hydropower concepts. This report outlines the game’s design philosophy, educational objectives, technical implementation, user experience insights, and potential for broader deployment within educational and workforce development contexts. It emphasizes how gamified learning can demystify complex hydropower science and inspire interest in water-power technologies. The project’s success demonstrates the value of Seedling funding in fostering creative, lowcost educational tools that support DOE’s mission to advance energy literacy and innovation. The report concludes with recommendations for expanding the tool by adding new levels, integrating assessment metrics, and exploring commercialization or deployment pathways through future Sapling funding. The official webpage of the Hydropower Game, which includes a link to the educational tool, can be accessed at www.anl.gov/hydropower/hydropower-game.

13 HYDRO ENERGY↗
NASA NTRS2008

Multiple-Instance Regression with Structured Data

We present a multiple-instance regression algorithm that models internal bag structure to identify the items most relevant to the bag labels. Multiple-instance regression (MIR) operates on a set of bags with real-valued labels, each containing a set of unlabeled items, in which the relevance of each item to its bag label is unknown. The goal is to predict the labels of new bags from their contents. Unlike previous MIR methods, MI-ClusterRegress can operate on bags that are structured in that they contain items drawn from a number of distinct (but unknown) distributions. MI-ClusterRegress simultaneously learns a model of the bag's internal structure, the relevance of each item, and a regression model that accurately predicts labels for new bags. We evaluated this approach on the challenging MIR problem of crop yield prediction from remote sensing data. MI-ClusterRegress provided predictions that were more accurate than those obtained with non-multiple-instance approaches or MIR methods that do not model the bag structure.

learning↗
DOE OSTI2026

Digital Twin Technology for Safety, Security, and Training in Spent Nuclear Fuel Handling

The increasing complexity of spent nuclear fuel handling requires significant resources to ensure safety, security, and personnel training. As nuclear facilities have continued to advance in scale and technology, the integration of digital tools has become indispensable. Among these tools, digital twins, which are virtual models of physical systems, are emerging as invaluable tools for enhancing safety protocols, security measures, and training in the nuclear sector. These models were conceptualized in the Industry 4.0 revolution. Digital twins can process data from physical systems in real time (by using sensors), include multiple code packages to enable simulations of different physics applications, and even implement artificial intelligence or machine learning techniques for advanced data processing. Despite the advantages that digital twins provide, challenges still exist regarding their widespread implementation. For instance, data used by a digital twin must be accurate to ensure that the digital twin is accurately tuned. Furthermore, if insecure digital twins are targeted by hackers, then they can pose serious risks to the security and safety of nuclear facilities.

Digital twins↗
NASA NTRS2004

A Decision-Theoretic Approach to Autonomous Planetary Rover Control

The report discusses the: Decentralized Control of Markov Decision Processes. Study the complexity of decentralized control of Markov decision processes, and develop algorithms for finding optimal control policies. Scheduling Contract Algorithms. Develop an optimal method for scheduling runs of a contract anytime algorithm (one that takes the deadline as input) in situations where the deadline is unknown, multiple problem instances must be solved, and a multi-processor machine is available. Planetary Rover Control as a Markov Decision Process.Use the Markov decision process framework to formalize and solve problems in planetary rover control. Adaptive Peer Selection. Use reinforcement learning to maximize the expected down-load speed for a client in a peer-to-peer file sharing system.

Zilberstein, Shlomo↗
DOE OSTI2024

SQMS Quantum R&D in Machine Learning, Optimization and Sensing beyond Fundamental Physics Applications

This newly formed team at SQMS under the Ecosystem Thrust is looking to develop capabilities impacting societal advances outside the core domain of HEP and condensed matter physics. We explicitly leverage the experimental and algorithmic innovations developed across all groups as well as connect to broad-scope external projects of the diverse team of PIs. As the inaugural set of projects, we are studying numerically quantum machine learning models inspired by efficiently trainable echo-state and orthogonal neural networks and developing designs for related experiments to be performed on quantum processors based on SQMS SRF cQED technology and Rigetti s transmon arrays. Investigated models exploit ideas and lessons learned from multiple prior work by SQMS team members in a variety of internal and external activities [R1]. Target initial applications include noisy signal processing, potentially captured by quantum sensors or noisy QPUs, as well as simulation and classification of healthcare data. For instance, image reconstruction of the brain s electrical properties by solving the inverse Maxwell equation problem with uncertainty [R2] through a hybrid quantum-classical physics-informed architecture for time-dependent processes [R3]. The group is also investigating the application and development of novel quantum sensors based on magnetic levitation of a superconducting sphere coupled to a superconducting qubit. This coupling enables high-precision measurements of the position of the sphere, which can be used for sensitive detection of forces, enabling practical applications such as gravimetry for geophysics analysis, or accelerometry for GPS-denied navigation [R4] [R1] Rieffel, Eleanor G., Ata Akbari Asanjan, M. Sohaib Alam, Namit Anand, David E. Bernal Neira, Sophie Block, Lucas T. Brady et al. "Assessing and advancing the potential of quantum computing: A NASA case study." Future Generation Computer Systems (2024). [R2] Yu, X., Serrall s, J.E., Giannakopoulos, I.I., Liu, Z., Daniel, L., Lattanzi, R. and Zhang, Z., 2023. Pifon-ept: Mr-based electrical property tomography using physics-informed fourier networks. IEEE Journal on Multiscale and Multiphysics Computational Techniques. [R3] Wudarski, Filip, Daniel OConnor, Shaun Geaney, Ata Akbari Asanjan, Max Wilson, Elena Strbac, P. Aaron Lott, and Davide Venturelli. "Hybrid quantum-classical reservoir computing for simulating chaotic systems." arXiv preprint arXiv:2311.14105 (2023). [R4] Higgins, Gerard, Saarik Kalia, and Zhen Liu. "Maglev for dark matter: Dark-photon and axion dark matter sensing with levitated superconductors." Physical Review D 109.5 (2024): 055024.

Venturelli, Davide↗
NASA NTRS2004

The Development of Two Science Investigator-led Processing Systems (SIPS) for NASA's Earth Observation System (EOS)

In 2001, NASA Goddard Space Flight Center's Laboratory for Terrestrial Physics started the construction of a science Investigator-led Processing System (SIPS) for processing data from the Ozone Monitoring Instrument (OMI) which will launch on the Aura platform in mid 2004. The Ozone Monitoring Instrument (OMI) is a contribution of the Netherlands Agency for Aerospace Programs (NIVR) in collaboration with the Finnish Meteorological Institute (FMI) to the Earth Observing System (EOS) Aura mission. It will continue the Total Ozone Monitoring System (TOMS) record for total ozone and other atmospheric parameters related to ozone chemistry and climate. OMI measurements will be highly synergistic with the other instruments on the EOS Aura platform. The LTP previously developed the Moderate Resolution Imaging Spectrometer (MODIS) Data Processing System (MODAPS), which has been in full operations since the launches of the Terra and Aqua spacecrafts in December, 1999 and May, 2002 respectively. During that time, it has continually evolved to better support the needs of the MODIS team. We now run multiple instances of the system managing faster than real time reprocessings of the data as well as continuing forward processing. The new OMI Data Processing System (OMIDAPS) was adapted from the MODAPS. It will ingest raw data from the satellite ground station and process it to produce calibrated, geolocated higher level data products. These data products will be transmitted to the Goddard Distributed Active Archive Center (GDAAC) instance of the Earth Observing System (EOS) Data and Information System (EOSDIS) for long term archive and distribution to the public. The OMIDAPS will also provide data distribution to the OMI Science Team for quality assessment, algorithm improvement, calibration, etc. We have taken advantage of lessons learned from the MODIS experience and software already developed for MODIS. We made some changes in the hardware system organization, database and software to adapt the system for OMI. We replaced the fundamental database system, Sybase, with an Open Source RDBMS called PostgreSQL, and based the entire OMIDAPS on a cluster of Linux based commodity computers rather than the large SGI servers that MODAPS uses. Rather than relying on a central I/O server host, the new system distributes its data archive among multiple server hosts in the cluster. OMI is also customizing the graphical user interfaces and reporting structure to more closely meet the needs of the OMI Science Team. Prior to 2003, simulated OMI data and the science algorithms were not ready for production testing. We initially constructed a prototype system and tested using a 25 year dataset of Total Ozone Mapping Spectrometer (TOMS) and Solar Backscatter Ultraviolet Instrument (SBUV) data. This prototype system provided a platform to support the adaptation of the algorithms for OMI, and provided reprocessing of the historical data aiding in its analysis. In a recent reanalysis of the TOMS data, the OMIDAPS processed 108,000 full orbits of data through 4 processing steps per orbit, producing about 800,000 files (400 GiB) of level 2 and greater data files. More recently we have installed two instances of the OMIDAPS for integration and testing of OM1 science processes as they get delivered from the Science Team. A Test instance of the OMIDAPS has also supported a series of "Interface Confidence Tests" (ICTs) and End-to-End Ground System tests to ensure the launch readiness of the system. This paper will discuss the high-level hardware, software, and database organization of the OMIDAPS and how it builds on the MODAPS heritage system. It will also provide an overview of the testing and implementation of the production OMIDAPS.

Tilmes, Curt↗
DOE OSTI2024

Toward Accelerating Discovery via Physics-Driven and Interactive Multifidelity Bayesian Optimization

Both computational and experimental material discovery bring forth the challenge of exploring multidimensional and often nondifferentiable parameter spaces, such as phase diagrams of Hamiltonians with multiple interactions, composition spaces of combinatorial libraries, processing spaces, and molecular embedding spaces. Often these systems are expensive or time consuming to evaluate a single instance, and hence classical approaches based on exhaustive grid or random search are too data intensive. This resulted in strong interest toward active learning methods such as Bayesian optimization (BO) where the adaptive exploration occurs based on human learning (discovery) objective. However, classical BO is based on a predefined optimization target, and policies balancing exploration and exploitation are purely data driven. In practical settings, the domain expert can pose prior knowledge of the system in the form of partially known physics laws and exploration policies often vary during the experiment. Here, we propose an interactive workflow building on multifidelity BO (MFBO), starting with classical (data-driven) MFBO, then expand to a proposed structured (physics-driven) structured MFBO (sMFBO), and finally extend it to allow human-in-the-loop interactive interactive MFBO (iMFBO) workflows for adaptive and domain expert aligned exploration. These approaches are demonstrated over highly nonsmooth multifidelity simulation data generated from an Ising model, considering spin–spin interaction as parameter space, lattice sizes as fidelity spaces, and the objective as maximizing heat capacity. Detailed analysis and comparison show the impact of physics knowledge injection and real-time human decisions for improved exploration with increased alignment to ground truth. Here, the associated notebooks allow to reproduce the reported analyses and apply them to other systems.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

BULKI-Store v0.3.2

BULKI-Store is a distributed object storage system optimized for high-performance computing environments. Built with a Rust core and Python bindings, it efficiently manages scientific and machine learning datasets across HPC clusters. The system employs a client-server architecture with MPI integration, enabling seamless scaling on supercomputers like Perlmutter. BULKI-Store's object-oriented approach provides intuitive data organization with rich metadata support, contrasting with traditional file-based solutions. Key optimizations include selective checkpoint loading, unified checkpoint files, and object chunking for large data transfers. For machine learning workloads, BULKI-Store offers advantages through fine-grained access patterns, dynamic data sharing between training instances, and reduced memory pressure. Memory management features include strategic Python GC calls, minimized data copies, and batch processing capabilities. The system leverages Rayon's thread pool for asynchronous data prefetching and supports multiple CPU architectures (ARM64, x86, AMD, RISC-V). By combining performance optimizations with developer-friendly APIs, BULKI-Store addresses the complex data management challenges of modern HPC applications while maintaining compatibility across heterogeneous computing environments.

Zhang, Wei [Lawrence Berkeley National Laboratory ↗
NASA NTRS1992

SkICAT: A cataloging and analysis tool for wide field imaging surveys

We describe an integrated system, SkICAT (Sky Image Cataloging and Analysis Tool), for the automated reduction and analysis of the Palomar Observatory-ST ScI Digitized Sky Survey. The Survey will consist of the complete digitization of the photographic Second Palomar Observatory Sky Survey (POSS-II) in three bands, comprising nearly three Terabytes of pixel data. SkICAT applies a combination of existing packages, including FOCAS for basic image detection and measurement and SAS for database management, as well as custom software, to the task of managing this wealth of data. One of the most novel aspects of the system is its method of object classification. Using state-of-theart machine learning classification techniques (GID3* and O-BTree), we have developed a powerful method for automatically distinguishing point sources from non-point sources and artifacts, achieving comparably accurate discrimination a full magnitude fainter than in previous Schmidt plate surveys. The learning algorithms produce decision trees for classification by examining instances of objects classified by eye on both plate and higher quality CCD data. The same techniques will be applied to perform higher-level object classification (e.g., of galaxy morphology) in the near future. Another key feature of the system is the facility to integrate the catalogs from multiple plates (and portions thereof) to construct a single catalog of uniform calibration and quality down to the faintest limits of the survey. SkICAT also provides a variety of data analysis and exploration tools for the scientific utilization of the resulting catalogs. We include initial results of applying this system to measure the counts and distribution of galaxies in two bands down to Bj is approximately 21 mag over an approximate 70 square degree multi-plate field from POSS-II. SkICAT is constructed in a modular and general fashion and should be readily adaptable to other large-scale imaging surveys.

Weir, N.↗
DOE OSTI2025

Low responsiveness of machine learning models to critical or deteriorating health conditions

Machine learning (ML) based mortality prediction models can be immensely useful in intensive care units. Such a model should generate warnings to alert physicians when a patient’s condition rapidly deteriorates, or their vitals are in highly abnormal ranges. Before clinical deployment, it is important to comprehensively assess a model’s ability to recognize critical patient conditions. We develop multiple medical ML testing approaches, including a gradient ascent method and neural activation map. We systematically assess these machine learning models’ ability to respond to serious medical conditions using additional test cases, some of which are time series. Guided by medical doctors, our evaluation involves multiple machine learning models, resampling techniques, and four datasets for two clinical prediction tasks. We identify serious deficiencies in the models’ responsiveness, with the models being unable to recognize severely impaired medical conditions or rapidly deteriorating health. For in-hospital mortality prediction, the models tested using our synthesized cases fail to recognize 66% of the injuries. In some instances, the models fail to generate adequate mortality risk scores for all test cases. Our study identifies similar kinds of deficiencies in the responsiveness of 5-year breast and lung cancer prediction models. Using generated test cases, we find that statistical machine-learning models trained solely from patient data are grossly insufficient and have many dangerous blind spots. Most of the ML models tested fail to respond adequately to critically ill patients. How to incorporate medical knowledge into clinical machine learning models is an important future research direction.

60 APPLIED LIFE SCIENCES↗