Search NASA⌕ Search

SEARCH · Search NASA

Results for “supervised training”

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 163 records · Page 9
All sourcesNASA NTRSDOE OSTI
DOE OSTI2020

Multiperspective Automotive Labeling

Supervised machine learning techniques inherently rely on datasets to be trained. With image datasets traditionally being annotated by humans, many advancements in image annotation tools have been made to ensure creation of rich datasets with accurate labels. Nevertheless, users still find it challenging to create and use their own datasets with labels that reflect their problem domain. We propose a streamlined labeling process that aligns multiperspective images and allows a transition from a labeled perspective to other perspectives. The main goal of this work is to reduce the human effort required for labeling vehicle images under favorable conditions where the image perspectives are correlated and one or more perspectives are known. A case study is described and analyzed to show the effectiveness of the process, as well as constraints and limitations when applied to other cases.

training dataset, multiperspective images, supervi↗
DOE OSTI2024

Data driven discovery and quantification of hyperspectral leaf reflectance phenotypes across a maize diversity panel

Abstract Estimates of plant traits derived from hyperspectral reflectance data have the potential to efficiently substitute for traits, which are time or labor intensive to manually score. Typical workflows for estimating plant traits from hyperspectral reflectance data employ supervised classification models that can require substantial ground truth datasets for training. We explore the potential of an unsupervised approach, autoencoders, to extract meaningful traits from plant hyperspectral reflectance data using measurements of the reflectance of 2151 individual wavelengths of light from the leaves of maize ( Zea mays ) plants harvested from 1658 field plots in a replicated field trial. A subset of autoencoder‐derived variables exhibited significant repeatability, indicating that a substantial proportion of the total variance in these variables was explained by difference between maize genotypes, while other autoencoder variables appear to capture variation resulting from changes in leaf reflectance between different batches of data collection. Several of the repeatable latent variables were significantly correlated with other traits scored from the same maize field experiment, including one autoencoder‐derived latent variable (LV8) that predicted plant chlorophyll content modestly better than a supervised model trained on the same data. In at least one case, genome‐wide association study hits for variation in autoencoder‐derived variables were proximal to genes with known or plausible links to leaf phenotypes expected to alter hyperspectral reflectance. In aggregate, these results suggest that an unsupervised, autoencoder‐based approach can identify meaningful and genetically controlled variation in high‐dimensional, high‐throughput phenotyping data and link identified variables back to known plant traits of interest.

Tross, Michael C.↗
DOE OSTI2026

U-net architected deep material network training with microstructure local field information

The Deep Material Network (DMN) has recently emerged as a powerful reduced-order modeling framework for simulating the mechanical response of heterogeneous materials such as composites. Unlike most data-driven approaches that directly learn a material’s response under prescribed loading, the DMN acts as a homogenization operator, learning the kinematic constraints and mechanical interactions of the underlying microstructure. However, traditional DMN training relies exclusively on homogenized effective properties derived from Direct Numerical Simulations (DNS), discarding the rich local field data that govern microstructural interactions. In this work, we extend the DMN framework to incorporate such local field information into the offline training process. Utilizing a U-Net architecture, we augment the DMN training objective to include the first and second statistical moments of the local stress fields obtained from linear DNS. This ensures that the learned network topology not only fits the effective stiffness but also accurately reflects the internal local stress and strain partitioning of the microstructure. The results confirm that supervising the localization process during training yields a superior surrogate model, reducing local prediction errors by an order of magnitude and significantly improving generalization to unseen nonlinear constitutive behaviors compared to traditional DMNs.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Evaluating the Use of Foundational Chemical Language Models in Multimodal Graph Fusion

Rapid and accurate prediction of the physicochemical properties of molecules given their structures remains a key challenge in cheminformatics. Machine learning approaches offer high-throughput options, but the optimality of inductive biases and data representations are up for debate. For example, BERT-based masked language models (MLMs) can be trained in a self-supervised way on hundreds of millions to billions of readily available SMILES strings. Another option is graph neural networks (GNNs), which can operate directly on molecular structures. Yet, generating accurate molecular geometry is computationally expensive, leading to a relative scarcity in data compared to SMILES strings. It is attractive to combine these two paradigms by pre-training an LM on a large corpus of SMILES strings and embedding these representation into a geometric graph neural network. Despite the promise of such an approach, and contrary to previous studies, we find mixed results with the combination of the LMs and GNNs on several molecule datasets. In particular, we found evidence for improvement on the FreeSolv and QM7 benchmarks, but degraded performance on the ESOL, LIPO and QM9 datasets compared to a GNN baseline.

Francel, Collin [University of Alabama]↗
DOE OSTI2024

Self-Supervised Cloud Classification

Abstract Low-level marine clouds play a pivotal role in Earth’s weather and climate through their interactions with radiation, heat and moisture transport, and the hydrological cycle. These interactions depend on a range of dynamical and microphysical processes that result in a broad diversity of cloud types and spatial structures, and a comprehensive understanding of cloud morphology is critical for continued improvement of our atmospheric modeling and prediction capabilities moving forward. Deep learning has recently accelerated our ability to study clouds using satellite remote sensing, and machine learning classifiers have enabled detailed studies of cloud morphology. A major limitation of deep learning approaches to this problem, however, is the large number of hand-labeled samples that are required for training. This work applies a recently developed self-supervised learning scheme to train a deep convolutional neural network (CNN) to map marine cloud imagery to vector embeddings that capture information about mesoscale cloud morphology and can be used for satellite image classification. The model is evaluated against existing cloud classification datasets and several use cases are demonstrated, including training cloud classifiers with very few labeled samples, interrogation of the CNN’s learned internal feature representations, cross-instrument application, and resilience against sensor calibration drift and changing scene brightness. The self-supervised approach learns meaningful internal representations of cloud structures and achieves comparable classification accuracy to supervised deep learning methods without the expense of creating large hand-annotated training datasets. Significance Statement Marine clouds heavily influence Earth’s weather and climate, and improved understanding of marine clouds is required to improve our atmospheric modeling capabilities and physical understanding of the atmosphere. Recently, deep learning has emerged as a powerful research tool that can be used to identify and study specific marine cloud types in the vast number of images collected by Earth-observing satellites. While powerful, these approaches require hand-labeling of training data, which is prohibitively time intensive. This study evaluates a recently developed self-supervised deep learning method that does not require human-labeled training data for processing images of clouds. We show that the trained algorithm performs competitively with algorithms trained on hand-labeled data for image classification tasks. We also discuss potential downstream uses and demonstrate some exciting features of the approach including application to multiple satellite instruments, resilience against changing image brightness, and its learned internal representations of cloud types. The self-supervised technique removes one of the major hurdles for applying deep learning to very large atmospheric datasets.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2021

Process Anomaly Detection for Sparsely Labeled Events in Nuclear Power Plants

An essential aspect of online monitoring, subtle anomaly detection increases the detection lead time for equipment failure and enables a nuclear power plant (NPP) to mitigate unexpected partial or full outages, resulting in significant cost saving to the plant. Once an anomaly is detected by plant staff, its cause and severity are investigated. Because the vast majority of anomalies require some level of investigation, including some that require time-consuming examination, before they are passed over to the engineering organization for further analysis, plants are often equipped with tools to assist the staff in performing anomaly detection. Those tools operate as a black box and are often based on statistical methods that establish sensor correlations using preconfigured mathematical models and flag correlation deviations as anomalies. Due to the number of anomalies detected at a given NPP on a daily basis, a significant number of flagged anomalies usually await examination for days or weeks. A primary cause of this backlog is that the methods used by the tools generate many false positives. Though this is usually attributed to oversensitive model settings due to very narrow normal operation bands, it can also be associated with the model development being inadequate for the process being monitored, or with missing model inputs that could have explained misclassified positives. The performance of anomaly detection tools impacts their plant acceptance and utilization, especially when the effort to address false positives generated by the tool depletes the value or cost saved by using that tool. Thus, means to advance anomaly detection performance have been investigated by the Department of Energy’s Light Water Reactor Sustainability program. Previous and ongoing efforts have targeted unsupervised machine-learning (ML) methods, which do not require the labeling of any data fed into the ML model. By contrast, in supervised anomaly detection methods, every data point is labeled as either a normal or abnormal process condition, and the model is trained to replicate the classification process. Supervised methods usually outperform unsupervised methods, due to the added value in differentiating normal from anomalous states of the monitored process. An NPP’s corrective action program requires it to track and document, via a dedicated report, the resolution of any issues that occur within the plant. Once created, each report is reviewed by a plant screening committee, and several classifications and decisions are made. Recently, a collaborating NPP developed an artificial intelligence and ML-based classifier to categorize a condition report (CR) into classes that can serve to label the data as normal or anomalous. Applying CRs as labels represents a semi-supervised use case. Semi-supervised ML assumes that labels exist for some data points (i.e., labeled anomalies, in this case) but not for the rest. In this effort, semi-supervised ML methods were used to fuse data from CRs with anomaly detection methods in order to test the hypothesis that partially labeled anomalies would improve the accuracy of the anomaly detection methods. Specifically, two methods were used. The first is the deep Semi-supervised Anomaly Detection (deep SAD) method, which can handle labels ranging from fully unsupervised to fully supervised cases. The second is a newly designed ML method developed specifically for this effort and referred to as the high-order feature (HOF)-based method. To evaluate these two methods in controlled environments, synthetic data generators were developed and used. The first datasets used a spring-mass-damper (SMD) system simulator commonly found in mechanical engineering references. This was used to create two use cases: a one- and a three-mass system. Anomalies were introduced by changing the spring and damper coefficients while the system was actuated by random forces. The second datasets used the commercial Dymola-Modelica software to build a simplified nuclear reactor model. Anomalies were added in the form of corrupted sensor readings and/or control commands. The deep SAD method was tested using the SMD system, while the HOF method was tested using both datasets. Application of the deep SAD semi-supervised ML method demonstrated that labels can generate increased confidence in detecting true anomalies. This helped increase the number of true positives and decrease the number of false negatives—something that would aid in addressing the backlog of possible anomalies. Application of the HOF method demonstrated that labels can aid in down selecting from a candidate set of features to a more optimal subset in order to better differentiate between normal and anomalous conditions.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2021

Training Restricted Boltzmann Machines With a D-Wave Quantum Annealer

Restricted Boltzmann Machine (RBM) is an energy-based, undirected graphical model. It is commonly used for unsupervised and supervised machine learning. Typically, RBM is trained using contrastive divergence (CD). However, training with CD is slow and does not estimate the exact gradient of the log-likelihood cost function. In this work, the model expectation of gradient learning for RBM has been calculated using a quantum annealer (D-Wave 2000Q), where obtaining samples is faster than Markov chain Monte Carlo (MCMC) used in CD. Training and classification results of RBM trained using quantum annealing are compared with the CD-based method. The performance of the two approaches is compared with respect to the classification accuracies, image reconstruction, and log-likelihood results. The classification accuracy results indicate comparable performances of the two methods. Image reconstruction and log-likelihood results show improved performance of the CD-based method. It is shown that the samples obtained from quantum annealer can be used to train an RBM on a 64-bit “bars and stripes” dataset with classification performance similar to an RBM trained with CD. Though training based on CD showed improved learning performance, training using a quantum annealer could be useful as it eliminates computationally expensive MCMC steps of CD.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2017

Semi-supervised learning approach for crack detection and identification in advanced gas-cooled reactor graphite bricks - 111

One of the life-limiting components of an Advanced Gas cooled Reactor (AGR) is its graphite core. The bricks present in the core undergo radiolytic oxidation throughout their lifetime which causes graphite weight loss and irradiation which can result in some of the bricks developing cracks. Understanding the nature and extent of brick cracking within the core is key to ensuring continued and extended operation of the AGR fleet. A semi-supervised machine learning classification algorithm is proposed as a method for improving the detection of cracked graphite bricks, by combining the labels derived from infrequent, detailed inspections of the core, with unlabeled, more frequent monitoring measurements taken during refueling operations. Semi-supervised machine learning, which is an emerging field in nuclear power condition monitoring, is the combination of ideas from both supervised and unsupervised machine learning whereby the data that is used to train the algorithm is a combination of labeled and unlabeled data. This paper introduces the initial research that has been undertaken in creating a semi-supervised self-training algorithm to detect the presence of graphite brick cracks and then proceeds to show that there is an improvement in the classification of graphite bricks using a semi-supervised machine learning classifier compared to supervised machine learning classifiers. This improved classification performance is encouraging as it does not require time consuming and costly human analysis to obtain extra learning information from available data. (authors)

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗
DOE OSTI2023

Semi-supervised graph neural networks for pileup noise removal

Abstract The high instantaneous luminosity of the CERN Large Hadron Collider leads to multiple proton–proton interactions in the same or nearby bunch crossings (pileup). Advanced pileup mitigation algorithms are designed to remove this noise from pileup particles and improve the performance of crucial physics observables. This study implements a semi-supervised graph neural network for particle-level pileup noise removal, by identifying individual particles produced from pileup. The graph neural network is firstly trained on charged particles with known labels, which can be obtained from detector measurements on data or simulation, and then inferred on neutral particles for which such labels are missing. This semi-supervised approach does not depend on the neutral particle pileup label information from simulation, and thus allows us to perform training directly on experimental data. The performance of this approach is found to be consistently better than widely-used domain algorithms and comparable to the fully-supervised training using simulation truth information. The study serves as the first attempt at applying semi-supervised learning techniques to pileup mitigation, and opens up a new direction of fully data-driven machine learning pileup mitigation studies.

43 PARTICLE ACCELERATORS↗
NASA NTRS1984

A comparison of Landsat point and rectangular field training sets for land-use classification

Rectangular training fields of homogeneous spectroreflectance are commonly used in supervised pattern recognition efforts. Trial image classification with manually selected training sets gives irregular and misleading results due to statistical bias. A self-verifying, grid-sampled training point approach is proposed as a more statistically valid feature extraction technique. A systematic pixel sampling network of every ninth row and ninth column efficiently replaced the full image scene with smaller statistical vectors which preserved the necessary characteristics for classification. The composite second- and third-order average classification accuracy of 50.1 percent for 331,776 pixels in the full image substantially agreed with the 51 percent value predicted by the grid-sampled, 4,100-point training set.

Tom, C. H.↗
DOE OSTI2024

SSTDR and FDR Detection of Un-Energized and Energized Cable Anomalies Including Thermal Degradation Using Machine Learning

Historically, cables are initially qualified for nuclear power plant use for 40 years. As plants extend their operating license to 60 and 80 years, continued use of these cables must shift to a performance-based approach since it is cost prohibitive to completely replace cables that are likely still capable of performing their design function. A variety of cable tests are available and are commonly applied during outages when the cables can be taken out of service. Frequency domain reflectometry (FDR) is one of these test methods that is being more broadly accepted and used because it not only detects anomalies along the cable with a low-voltage signal that does not stress the cable insulation, but the technique also locates the anomalies. This supports follow-up local inspection and local repair or partial replacement of a damaged cable segment. Currently, FDR testing is only applied to cables that are taken out of service since the test instrument would be damaged by operational voltages. A related technology that has found some acceptance in the aircraft and rail industry is spread spectrum time domain reflectometry (SSTDR). This technology has been implemented with a custom commercial instrument by LiveWire Innovation that is designed to operate on live cables up to 1000 volts and with a bandwidth of 48 MHz. Initial evaluation by the Pacific Northwest National Laboratory (PNNL) of the Live Wire system indicated that a broader bandwidth (BW) SSTDR may be better for many kinds of flaws. This led PNNL to develop an SSTDR laboratory instrument suitable for tests up to 500 MHz bandwidth. Testing on energized cables is also desirable for online monitoring systems so an inductive clamshell coupler was developed that allows energized cables to be tested up to at least 5 kV and likely higher voltage levels. Dielectric spectroscopy and tan delta testing plus various laboratory destructive tests were included in this data acquisition campaign directed to feed a machine learning (ML) study. With these kinds of developments, online energized cable tests may be possible with industrial adoption of such hardware advances but it will be completely impractical to have highly skilled data analysts continually examine these complex signals for indications of damage or compromised conditions. If online testing is to be implemented in new test hardware, it must be accompanied by software that can interpret the signals and alert plant operators of changing or degraded conditions. The thermally aged, shielded cable investigated here was separately treated for ML analysis. Visual analysis of electrical data showed generally increasing peaks where the cable entered and exited the oven. These peaks were not exactly aligned with expected locations, but these differences were attributed to velocity of propagation calibration errors. Only supervised ML was applied to the thermally aged data as this data was only available shortly before the committed publication date of this report. The supervised ML was structured to divide the 0 to 70-day responses as ‘normal’ from 0 to 35 days or ‘anomalous’ from 36 to 70 days, based on cable tensile elongation at break (EAB) insulation characterization. Using 80% of the data for training and 20% for testing, the supervised ML predicted normal versus anomalous was 70% accurate. Important conclusions include: • Accuracy to predict the presence of cable damage is improved from the 2023 effort by more training data. Weighted accuracies for comparisons among the instruments ranged from 67 to 89 % for unsupervised ML and 71 to 99% for supervised ML. • Based on the synthetic data tests, the unsupervised models are more generalizable to unseen anomalies. The Multi-Layer Perceptron classifier (MLP) model reported as high as 99.7% accuracy on the test data, but this dropped to 58.3% when tested on the synthetic data. In contrast, the unsupervised Pointwise model only achieved 89.7% accuracy on the experimental data but reported 78.3% accuracy on the synthetic data. • The best anomaly indicators are higher frequency (400 MHz BW) FDR data. Other tests may be interesting but for this study, this was the best predicter.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2023

Spread Spectrum Time Domain Reflectometry (SSTDR) and Frequency Domain Reflectometry (FDR) for Detection of Cable Anomalies Using Machine Learning

Cables are initially qualified for nuclear power plant use for 40 years. As plants extend their operating license to 60 and 80 years, continued use of these cables must shift to a performance-based approach since it is cost prohibitive to completely replace cables that are likely still capable of performing their design function. A variety of cable tests are available and are commonly applied during outages when the cables can be taken out of service. Frequency domain reflectometry (FDR) is one of these test methods that is being more broadly accepted and used because it not only detects anomalies along the cable with a low-voltage signal that does not stress the cable insulation, but the technique also locates the anomalies. This supports follow-up local inspection and local repair or partial replacement of a damaged cable segment. Currently, FDR testing is only applied to cables that are taken out of service since the test instrument would be damaged by operational voltages. A related technology that has found some acceptance in the aircraft and rail industry is spread spectrum time domain reflectometry (SSTDR). This technology has been implemented with a custom commercial instrument by LiveWire Innovation Inc. that is designed to operate on live cables up to 1000 volts. One of the main conclusions of a previous effort was that cable reflectometry plots can be difficult for humans to analyze due to baseline noise, low or noisy anomaly response peaks, or large responses from cable ends. Detection of cable anomalies for many of these frequencies and test conditions was challenging for manual analysis. This presented an ideal opportunity for ML analysis to distinguish undamaged cable indications from anomalous cable indications. This research discusses application of machine learning (ML) to reflectometry cable test methods. The goal was to assess feasibility to distinguish undamaged cable reflectometry responses from damaged or anomalous cable reflectometry responses. The assessment considered the 3 instruments, multiple frequency bandwidths from each instrument, multiple cable anomalies and test conditions, and both supervised and unsupervised ML approaches. Although approaches and analysis methods were not identical or directly comparable, both outputs were encouraging. The unsupervised prediction weighted accuracy was assessed by instrument and by frequency. It performed better at high frequencies with the highest prediction accuracy of 0.84 for the higher frequency FDR, 0.79 for the 48-MHz LiveWire SSTDR, and 0.77 for 300-MHz PNNL SSTDR. The initial weighted accuracy average across all frequencies for using supervised ML was 0.56 to 0.68. The supervised analysis was repeated with noisier training data removed resulting in weighted accuracies of 0.69 to 0.87. These weighted accuracies are not directly comparable due to differences in the supervised and unsupervised analysis details but do indicate an encouraging trend. Even with limited and unbalanced data, strong prediction accuracies seem encouraging for further work including more data under a wider range of conditions.

42 ENGINEERING↗
DOE OSTI2022

Photodiode-based machine learning for optimization of laser powder bed fusion parameters in complex geometries

We report the quality of parts produced through laser powder bed fusion additive manufacturing can be irregular, with complex geometries sometimes exhibiting dimensional inaccuracies and defects. For optimal part quality, laser process parameters should be selected carefully prior to printing and adjusted during the print if necessary. This is challenging since approaches to control and optimize the build parameters need to take into account the part geometry, the material, and the complex physics of laser powder bed fusion. This work describes a data-driven approach using experimental diagnostics for the optimization of laser process parameters prior to printing. A training dataset is generated by collecting high speed photodiode signal data while printing simple parts containing key geometry features with various process parameter strategies. Supervised learning approaches are employed to train both a forward model and an inverse model. The forward model takes as inputs track-wise geometry features and laser parameters and outputs the photodiode signal along the scan path. The inverse model takes as inputs the geometry features and photodiode signal and predicts the laser parameters. Given the part geometry and a desired photodiode signal, the inverse model can thus determine the required laser parameters. Two test parts which contain defect-prone features are used to assess the validity of the inverse model. The use of the model leads to improved part quality (higher dimensional accuracy, reduced dross, reduced distortion) for both test geometries.

36 MATERIALS SCIENCE↗
DOE OSTI2022

The Power of Priors: Improved Enrichment Safeguards

International safeguards currently rely on material accountancy to verify that declared nuclear material is present and unmodified. Although effective, material accountancy for large bulk facilities can be expensive to implement due to the high precision instrumentation required to meet regulatory targets. Process monitoring has long been considered to improve material accountancy. However, effective integration of process monitoring has been met with mixed results. Given the large successes in other domains, machine learning may present a solution for process monitoring integration. Past work has shown that unsupervised approaches struggle due to measurement error. Although not studied in depth for a safeguards context, supervised approaches often have poor generalization for unseen classes of data (e.g., unseen material loss patterns). This work shows that engineered datasets, when used for training, can improve the generalization of supervised approaches. Further, the underlying models needed to generate these datasets need only accurately model certain high importance features.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
DOE OSTI2012

Literature mining of protein-residue associations with graph rules learned through distant supervision

Background: We propose a method for automatic extraction of protein-specific residue mentions from the biomedical literature. The method searches text for mentions of amino acids at specific sequence positions and attempts to correctly associate each mention with a protein also named in the text. The methods presented in this work will enable improved protein functional site extraction from articles, ultimately supporting protein function prediction. Our method made use of linguistic patterns for identifying the amino acid residue mentions in text. Further, we applied an automated graph-based method to learn syntactic patterns corresponding to protein-residue pairs mentioned in the text. We finally present an approach to automated construction of relevant training and test data using the distant supervision model. Results: The performance of the method was assessed by extracting protein-residue relations from a new automatically generated test set of sentences containing high confidence examples found using distant supervision. It achieved a F-measure of 0.84 on automatically created silver corpus and 0.79 on a manually annotated gold data set for this task, outperforming previous methods. Conclusions: The primary contributions of this work are to (1) demonstrate the effectiveness of distant supervision for automatic creation of training data for protein-residue relation extraction, substantially reducing the effort and time involved in manual annotation of a data set and (2) show that the graph-based relation extraction approach we used generalizes well to the problem of protein-residue association extraction. This work paves the way towards effective extraction of protein functional residues from the literature.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2022

Event‐Based Training in Label‐Limited Regimes

Abstract The distribution of attributes assigned using data on independent sensors for a specific source, for example, magnitude, can be richly descriptive for final event characterization and associated uncertainty. Attribute distributions can also provide powerful context for event characterization in the absence of comprehensive annotation. This work develops a way to leverage distributional information across a set of sensors in the absence of comprehensive annotation as a domain‐informed regularization term applied during gradient‐based learning. The regularization term is the basis of event‐based training which I show can be a powerful semi‐supervised learning (SSL) approach. I first use a simple feed forward neural network and a toy data set to outline how data set structure interacts with the assumptions inherent to many semi‐supervised learning approaches. I then demonstrate the effectiveness of event‐based training using a deep convolutional neural network for seismic event classification in Utah, which increases SSL accuracy from 92% to 97% on event classification with a limited number of training labels.

Linville, Lisa M.↗
DOE OSTI2024

TCR-H: explainable machine learning prediction of T-cell receptor epitope binding on unseen datasets

Artificial-intelligence and machine-learning (AI/ML) approaches to predicting T-cell receptor (TCR)-epitope specificity achieve high performance metrics on test datasets which include sequences that are also part of the training set but fail to generalize to test sets consisting of epitopes and TCRs that are absent from the training set, i.e., are ‘unseen’ during training of the ML model. We present TCR-H, a supervised classification Support Vector Machines model using physicochemical features trained on the largest dataset available to date using only experimentally validated non-binders as negative datapoints. TCR-H exhibits an area under the curve of the receiver-operator characteristic (AUC of ROC) of 0.87 for epitope ‘hard splitting’ (i.e., on test sets with all epitopes unseen during ML training), 0.92 for TCR hard splitting and 0.89 for ‘strict splitting’ in which neither the epitopes nor the TCRs in the test set are seen in the training data. Furthermore, we employ the SHAP (Shapley additive explanations) eXplainable AI (XAI) method for post hoc interrogation to interpret the models trained with different hard splits, shedding light on the key physiochemical features driving model predictions. TCR-H thus represents a significant step towards general applicability and explainability of epitope:TCR specificity prediction.

60 APPLIED LIFE SCIENCES↗