SEARCH · Engineering Papers

Results for “latent feature models”

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.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 73 records · Page 4

All sources NASA NTRS DOE OSTI

DOE OSTI2026

Deep generative learning of magnetic frustration in artificial spin ice from magnetic force microscopy images

Increasingly large datasets of microscopic images with nanoscale resolution facilitate the development of machine learning methods to identify and analyze subtle physical phenomena embedded within the images. In this work, microscopic images of honeycomb lattice spin-ice samples serve as datasets from which we automate the calculation of net magnetic moments and directional orientations of spin-ice configurations. In the first stage of our workflow, machine learning models are trained to accurately predict magnetic moments and directions within spin-ice structures. Variational Autoencoders (VAEs), an emergent unsupervised deep learning technique, are employed to generate high-quality synthetic magnetic force microscopy (MFM) images and extract latent feature representations, thereby reducing experimental and segmentation errors. The second stage of proposed methodology enables precise identification and prediction of frustrated vertices and nanomagnetic segments, effectively correlating structural and functional aspects of microscopic images. This facilitates the design of optimized spin-ice configurations with controlled frustration patterns, enabling potential on-demand synthesis.

36 MATERIALS SCIENCE↗

DOE OSTI2024

CO 2 storage site characterization using ensemble-based approaches with deep generative models

Estimating spatially distributed properties such as permeability from available sparse measurements is a great challenge in efficient subsurface CO 2 storage operations. In this paper, a deep generative model that can accurately capture complex subsurface structure is tested with an ensemble-based inversion method for accurate and accelerated characterization of CO 2 storage sites. We chose Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) for its realistic reservoir property representation and Ensemble Smoother with Multiple Data Assimilation (ES-MDA) for its robust data fitting and uncertainty quantification capability. WGAN-GP are trained to generate high-dimensional permeability fields from a low-dimensional latent space and ES-MDA then updates the latent variables by assimilating available measurements. Several subsurface site characterization examples including Gaussian, channelized, and fractured reservoirs are used to evaluate the accuracy and computational efficiency of the proposed method and the main features of the unknown permeability fields are characterized accurately with reliable uncertainty quantification. Furthermore, the estimation performance is compared with a widely-used variational, i.e., optimization-based, inversion approach, and the proposed approach outperforms the variational inversion method in several benchmark cases. We explain such superior performance by visualizing the objective function in the latent space: because of nonlinear and aggressive dimension reduction via generative modeling, the objective function surface becomes extremely complex while the ensemble approximation can smooth out the multi-modal surface during the minimization. This suggests that the ensemble-based approach works well over the variational approach when combined with deep generative models at the cost of forward model runs unless convergence-ensuring modifications are implemented in the variational inversion.

42 ENGINEERING↗

DOE OSTI2021

Self-Supervised Anomaly Detection via Neural Autoregressive Flows with Active Learning

Many self-supervised methods have been proposed with the target of image anomaly detection. These methods often rely on the paradigm of data augmentation with predefined transformations such as flipping, cropping, and rotations. However, it is not straightforward to apply these techniques for non-image data, such as time series or tabular data, while the performance of the existing deep approaches has been under our expectation on tasks beyond images. In this work, we propose a novel active learning (AL) scheme that relied on neural autoregressive flows (NAF) for self-supervised anomaly detection, specifically on small-scale data. Unlike other generative models such as GANs or VAEs, flow-based models allow to explicitly learn the probability density and thus can assign accurate likelihoods to normal data which makes it usable to detect anomalies. The proposed NAF-AL method is achieved by efficiently generating random samples from latent space and transforming them into feature space along with likelihoods via invertible mapping. The samples with lower likelihoods are selected and further checked by outlier detection using Mahalanobis distance. The augmented samples incorporating with normal samples are used for training a better detector so as to approach decision boundaries. Compared with random transformations, NAF-AL can be interpreted as a likelihood-oriented data augmentation that is more efficient and robust. Extensive experiments show that our approach outperforms existing baselines on multiple time series and tabular datasets, and a real-world application in advanced manufacturing, with significant improvement on anomaly detection accuracy and robustness over the state-of-the-art.

Zhang, Jiaxin↗

DOE OSTI2023

A Graphical Model for Fusing Diverse Microbiome Data

This paper develops a Bayesian graphical model for fusing disparate types of count data. The motivating application is the study of bacterial communities from diverse high-dimensional features, in this case, transcripts, collected from different treatments. In such datasets, there are no explicit correspondences between the communities and each corresponds to different factors, making data fusion challenging. We introduce a flexible multinomial-Gaussian generative model for jointly modeling such count data. This latent variable model jointly characterizes the observed data through a common multivariate Gaussian latent space that parameterizes the set of multinomial probabilities of the transcriptome counts. The covariance matrix of the latent variables induces a covariance matrix of co-dependencies between all the transcripts, effectively fusing multiple data sources. We present a computationally scalable variational Expectation-Maximization (EM) algorithm for inferring the latent variables and the parameters of the model. Here, the inferred latent variables provide a common dimensionality reduction for visualizing the data and the inferred parameters provide a predictive posterior distribution. In addition to simulation studies that demonstrate the variational EM procedure, we apply our model to a bacterial microbiome dataset.

59 BASIC BIOLOGICAL SCIENCES↗

DOE OSTI2021

Characterizing Signatures of Geothermal Exploration Data with Machine Learning Techniques: An Application to the Nevada Play Fairway Analysis

We are introducing machine learning methods to the play fairway analysis to generate geothermal potential maps to support the evaluation of geothermal resource potential and the exploration for undiscovered blind geothermal systems in the Nevada Great Basin region. Our project aims to identify new ways to combine the play fairway data and empirically organize relationships between feature weights and labels in an improved workflow. As a means of doing this, we introduce machine learning methods to evaluate the influence of certain geological and geophysical features/feature sets in predicting geothermal favorability. This report highlights promising approaches based on supervised and unsupervised learning methods. First, we demonstrate a filter method applied to supervised classification modeling. The supervised filter method is based on permutation analysis to evaluate every possible feature combination/drop out scenario and rank feature influence based on the performance variance of supervised classification models. Additionally, we present an unsupervised factor analysis based on principal component analysis coupled with a semi-supervised kmeans clustering algorithm. This analysis allows us to identify the optimal number of groups/clusters for training sites and structural settings to identify feature patterns including correlation, variance, and latent and dominant feature relationships. The results from these methods offer a promising avenue for identifying favorable sources of predictive information to identify the locations of blind geothermal systems and furthering our understanding of complex geothermal feature and label relationships in the Great Basin region and beyond.

15 GEOTHERMAL ENERGY↗

DOE OSTI2024

Selecting Appropriate Model Complexity: An Example of Tracer Inversion for Thermal Prediction in Enhanced Geothermal Systems

Abstract A major challenge in the inversion of subsurface parameters is the ill‐posedness issue caused by the inherent subsurface complexities and the generally spatially sparse data. Appropriate simplifications of inversion models are thus necessary to make the inversion process tractable and meanwhile preserve the predictive ability of the inversion results. In this study, we investigate the effect of model complexity on fracture aperture inversion and thermal performance prediction in a field‐scale EGS model. Principal component analysis was used to map the aperture field to a low‐dimensional latent space. The complexity of the inversion model was quantitatively represented by the percentage of total variance in the original aperture fields preserved by the latent space. Tracer, pressure and flow rate data were used to invert for fracture aperture through an ensemble‐based inversion method, and the inferred aperture field was used to predict thermal performance. With an over‐simplified aperture model, ensemble collapse occurred. The inverted aperture models failed to resolve necessary flow and transport features, leading to a biased thermal performance prediction. A complex aperture model involved excessive features and was prone to overinterpreting the inversion data. Both the tracer/pressure/flow rate data reproduction and thermal prediction showed significant uncertainties, making it difficult to properly estimate long‐term thermal performance. Fortunately, our results indicate that there exists an appropriate model complexity which can simultaneously match inversion data and predict thermal performance with an acceptable uncertainty. The quality of the fit of tracer data appears to be a useful indicator of such an appropriate model complexity.

15 GEOTHERMAL ENERGY↗

DOE OSTI2026

CTGAN-TVAE

SAND2026-18914O CTGAN-TVAE (Conditional Tabular Generative Adversarial Networks-Tabular Variational Autoencoders) generates extensive sets of variable generation data through a hybrid framework. It enhances latent space representation by combining TVAE's robust feature-embedding with CTGAN's ability to condition categorical variables such as time. CTGAN-TVAE employs a fully connected neural network within a conditional generative adversarial network framework to manage continuous and categorical data effectively, capturing complex feature interactions without needing sequential modeling. This was developed as part of NNSA-MSIPP: Minority Serving Institution Partnership Program, Grant Number DE-NA0004016. Sandia National Laboratories is a multimission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy's National Nuclear Security Administration under contract DE-NA0003525.

Newlun, Cody [Sandia National Lab. (SNL-CA), Liver↗

DOE OSTI2023

Bridging the Gap between Cosmological Simulations with Graph Neural Networks and Domain Adaptation

Deep learning models have been shown to outperform methods that rely on summary statistics, like the power spectrum, in extracting information from complex cosmological data sets. However, due to differences in the subgrid physics implementation and numerical approximations across different simulation suites, models trained on data from one cosmological simulation show a drop in performance when tested on another. Similarly, models trained on any of the simulations would also likely experience a drop in performance when applied to observational data. Training on data from two different suites of the CAMELS hydrodynamic cosmological simulations, we examine the generalization capabilities of Domain Adaptive Graph Neural Networks (DA-GNNs). By utilizing GNNs, we capitalize on their capacity to capture structured scale-free cosmological information from galaxy distributions. Moreover, by including unsupervised domain adaptation via Maximum Mean Discrepancy (MMD), we enable our models to extract domain-invariant features. We demonstrate that DA-GNN achieves higher accuracy and robustness on cross dataset tasks (up to 28% better relative error and up to almost an order of magnitude better χ 2 ). Using data visualizations, we show the effects of domain adaptation on proper latent space data alignment. This shows that DA-GNNs are a promising method for extracting domain-independent cosmological information, a vital step toward robust deep learning for real cosmic survey data.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2023

Domain Adaptive Graph Neural Networks for Constraining Cosmological Parameters Across Multiple Data Sets

Deep learning models have been shown to outperform methods that rely on summary statistics, like the power spectrum, in extracting information from complex cosmological data sets. However, due to differences in the subgrid physics implementation and numerical approximations across different simulation suites, models trained on data from one cosmological simulation show a drop in performance when tested on another. Similarly, models trained on any of the simulations would also likely experience a drop in performance when applied to observational data. Training on data from two different suites of the CAMELS hydrodynamic cosmological simulations, we examine the generalization capabilities of Domain Adaptive Graph Neural Networks (DA-GNNs). By utilizing GNNs, we capitalize on their capacity to capture structured scale-free cosmological information from galaxy distributions. Moreover, by including unsupervised domain adaptation via Maximum Mean Discrepancy (MMD), we enable our models to extract domain-invariant features. We demonstrate that DA-GNN achieves higher accuracy and robustness on cross-dataset tasks (up to $28\%$ better relative error and up to almost an order of magnitude better $\chi^2$). Using data visualizations, we show the effects of domain adaptation on proper latent space data alignment. This shows that DA-GNNs are a promising method for extracting domain-independent cosmological information, a vital step toward robust deep learning for real cosmic survey data.

79 ASTRONOMY AND ASTROPHYSICS↗

DOE OSTI2024

Deep kernel methods learn better: from cards to process optimization

Abstract The ability of deep learning methods to perform classification and regression tasks relies heavily on their capacity to uncover manifolds in high-dimensional data spaces and project them into low-dimensional representation spaces. In this study, we investigate the structure and character of the manifolds generated by classical variational autoencoder (VAE) approaches and deep kernel learning (DKL). In the former case, the structure of the latent space is determined by the properties of the input data alone, while in the latter, the latent manifold forms as a result of an active learning process that balances the data distribution and target functionalities. We show that DKL with active learning can produce a more compact and smooth latent space which is more conducive to optimization compared to previously reported methods, such as the VAE. We demonstrate this behavior using a simple cards dataset and extend it to the optimization of domain-generated trajectories in physical systems. Our findings suggest that latent manifolds constructed through active learning have a more beneficial structure for optimization problems, especially in feature-rich target-poor scenarios that are common in domain sciences, such as materials synthesis, energy storage, and molecular discovery. The Jupyter Notebooks that encapsulate the complete analysis accompany the article.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2021

Evolution of Idealized Vortices in Monsoon-Like Shears: Application to Monsoon Depressions

Abstract This study examines processes fundamental to the development of South Asian monsoon depressions using an array of integrations of an idealized convection-permitting numerical model. In each integration, a wave of initially small amplitude is subjected to a different amount of vertical and meridional wind shear, with temperature and moisture fields constructed according to realistic constraints. Based on the evolution of this disturbance into monsoon depression–like vortices, two features of the background environment emerge as important: the low-level gradient of moist static energy (MSE) and the low-level meridional shear. As the low-level MSE gradient steepens, the disturbance becomes stronger and produces more rain. This strengthening results from the interaction of the vortex with latent heat release by convection that is in turn organized by positive MSE advection in the northerly flow west of the vortex. In this region of advection, moister air from the north ascends along upward-sloping isentropes, driving moist convection. The disturbance also becomes stronger with increasing meridional shear, which makes the environment more barotropically unstable. The absence of either of these two features of the background environment prevents substantial growth of the disturbance. Our results suggest that monsoon depression growth in South Asia is fostered by the coexistence of a strong low-level MSE gradient with strong meridional wind shear associated with the monsoon trough.

54 ENVIRONMENTAL SCIENCES↗

DOE OSTI2022

Dynamic logistic regression and variable selection: Forecasting and contextualizing civil unrest

Civil unrest can range from peaceful protest to violent furor, and researchers are working to monitor, forecast, and assess such events to allocate resources better. Twitter has become a real-time data source for forecasting civil unrest because millions of people use the platform as a social outlet. Additionally, daily word counts are used as model features, and predictive terms contextualize the reasons for the protest. To forecast civil unrest and infer the reasons for the protest, we consider the problem of Bayesian variable selection for the dynamic logistic regression model and propose using penalized credible regions to select parameters of the updated state vector. This method avoids the need for shrinkage priors, is scalable to high-dimensional dynamic data, and allows the importance of variables to vary in time as new information becomes available. A substantial improvement in both precision and F1-score using this approach is demonstrated through simulation. Finally, we apply the proposed model fitting and variable selection methodology to the problem of forecasting civil unrest in Latin America. Our dynamic logistic regression approach shows improved accuracy compared to the static approach currently used in event prediction and feature selection.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2023

Finding simplicity: unsupervised discovery of features, patterns, and order parameters via shift-invariant variational autoencoders *

Abstract Recent advances in scanning tunneling and transmission electron microscopies (STM and STEM) have allowed routine generation of large volumes of imaging data containing information on the structure and functionality of materials. The experimental data sets contain signatures of long-range phenomena such as physical order parameter fields, polarization, and strain gradients in STEM, or standing electronic waves and carrier-mediated exchange interactions in STM, all superimposed onto scanning system distortions and gradual changes of contrast due to drift and/or mis-tilt effects. Correspondingly, while the human eye can readily identify certain patterns in the images such as lattice periodicities, repeating structural elements, or microstructures, their automatic extraction and classification are highly non-trivial and universal pathways to accomplish such analyses are absent. We pose that the most distinctive elements of the patterns observed in STM and (S)TEM images are similarity and (almost-) periodicity, behaviors stemming directly from the parsimony of elementary atomic structures, superimposed on the gradual changes reflective of order parameter distributions. However, the discovery of these elements via global Fourier methods is non-trivial due to variability and lack of ideal discrete translation symmetry. To address this problem, we explore the shift-invariant variational autoencoders (shift-VAEs) that allow disentangling characteristic repeating features in the images, their variations, and shifts that inevitably occur when randomly sampling the image space. Shift-VAEs balance the uncertainty in the position of the object of interest with the uncertainty in shape reconstruction. This approach is illustrated for model 1D data, and further extended to synthetic and experimental STM and STEM 2D data. We further introduce an approach for training shift-VAEs that allows finding the latent variables that comport to known physical behavior. In this specific case, the condition is that the latent variable maps should be smooth on the length scale of the atomic lattice (as expected for physical order parameters), but other conditions can be imposed. The opportunities and limitations of the shift VAE analysis for pattern discovery are elucidated.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2021

Discovering Hidden Geothermal Signatures using Unsupervised Machine Learning

Discovering hidden geothermal resources is a very challenging task. It requires the mining of large datasets, including various diverse data attributes representing subsurface hydrogeological and geothermal conditions. The commonly used Play Fairway Analysis (PFA) typically relies on subject-matter expertise to analyze site or regional data to estimate geothermal conditions and prospectivity. Here, we demonstrate an alternative approach based on machine learning (ML) to process a geothermal dataset of Southwest New Mexico (SWNM). The study region includes low- and medium-temperature hydrothermal systems. However, most of these systems are poorly characterized because of insufficient existing data and limited past explorative studies. This study aims to discover hidden patterns and relationships in the SWNM geothermal dataset to better understand regional hydrothermal conditions. This is achieved by applying an unsupervised machine learning algorithm based on non-negative matrix factorization coupled with customized k-means clustering (NMFk). NMFk can automatically identify (1) hidden (latent) signatures characterizing datasets, (2) the optimal number of these signatures, (3) dominant data attributes associated with each signature, and (4) spatial distribution of the extracted signatures. Here, NMFk is applied to analyze 18 geological, geophysical, hydrogeological, geothermal attributes at 44 locations in SWNM. NMFk successfully finds data patterns and identifies the spatial associations of hydrothermal signatures with the four physiographic provinces in SWNM (Colorado Plateau, Volcanic Field, Basin and Range, and the Rio Grande rift). The algorithm identified up to 5 hydrothermal signatures in the SWNM datasets that differentiate between low- and medium-temperature hydrothermal systems in different provinces. Also, the algorithm identifies two medium-temperature hydrothermal systems in SWNM that require further exploration for geothermal resource development. Based on our analyses, 12 of the attributes are important to identify medium-temperature hydrothermal systems, and the remaining six attributes are critical to characterize low-temperature hydrothermal systems. Based on the obtained results, we identify potential physiographic provinces for further exploration to characterize them as geothermal resources. The resulting NMFk model can be applied to predict geothermal conditions and their uncertainties at new SWNM locations based on limited data from unexplored areas.

58 GEOSCIENCES↗

DOE OSTI2022

Estimation of the Surface Fluxes for Heat and Momentum in Unstable Conditions with Machine Learning and Similarity Approaches for the LAFE Data Set

Abstract Measurements of three flux towers operated during the land atmosphere feedback experiment (LAFE) are used to investigate relationships between surface fluxes and variables of the land–atmosphere system. We study these relations by means of two machine learning (ML) techniques: multilayer perceptrons (MLP) and extreme gradient boosting (XGB). We compare their flux derivation performance with Monin–Obukhov similarity theory (MOST) and a similarity relationship using the bulk Richardson number (BRN). The ML approaches outperform MOST and BRN. Best agreement with the observations is achieved for the friction velocity. For the sensible heat flux and even more so for the latent heat flux, MOST and BRN deviate from the observations while MLP and XGB yield more accurate predictions. Using MOST and BRN for latent heat flux, the root mean square errors (RMSE) are 107 Wm $$^{-2}$$ - 2 and 121 Wm $$^{-2}$$ - 2 , respectively, as well as the intercepts of the regression lines are $$\approx 110$$ ≈ 110 Wm $$^{-2}$$ - 2 . For the ML methods, the RMSEs reduce to 31 Wm $$^{-2}$$ - 2 for MLP and 33 Wm $$^{-2}$$ - 2 for XGB as well as the intercepts to just 4 Wm $$^{-2}$$ - 2 for MLP and $$-1$$ - 1 Wm $$^{-2}$$ - 2 for XGB with slopes of the regression lines close to 1, respectively. These results indicate significant deficiencies of MOST and BRN, particularly for the derivation of the latent heat flux. In fact, in contrast to the established theories, feature importance weighting demonstrates that the ML methods base their improved derivations on net radiation, the incoming and outgoing shortwave radiations, the air temperature gradient, and the available water contents, but not on the water vapor gradient. The results imply that further studies of surface fluxes and other turbulent variables with ML techniques provide great promise for deriving advanced flux parameterizations and their implementation in land–atmosphere system models.

54 ENVIRONMENTAL SCIENCES↗

DOE OSTI2021

Explaining and predicting human behavior and social dynamics in simulated virtual worlds: reproducibility, generalizability, and robustness of causal discovery methods

Ground Truth program was designed to evaluate social science modeling approaches using simulation test beds with ground truth intentionally and systematically embedded to understand and model complex Human Domain systems and their dynamics Lazer et al. (Science 369:1060–1062, 2020). Our multidisciplinary team of data scientists, statisticians, experts in Artificial Intelligence (AI) and visual analytics had a unique role on the program to investigate accuracy, reproducibility, generalizability, and robustness of the state-of-the-art (SOTA) causal structure learning approaches applied to fully observed and sampled simulated data across virtual worlds. In addition, we analyzed the feasibility of using machine learning models to predict future social behavior with and without causal knowledge explicitly embedded. In this paper, we first present our causal modeling approach to discover the causal structure of four virtual worlds produced by the simulation teams—Urban Life, Financial Governance, Disaster and Geopolitical Conflict. Our approach adapts the state-of-the-art causal discovery (including ensemble models), machine learning, data analytics, and visualization techniques to allow a human-machine team to reverse-engineer the true causal relations from sampled and fully observed data. We next present our reproducibility analysis of two research methods team’s performance using a range of causal discovery models applied to both sampled and fully observed data, and analyze their effectiveness and limitations. We further investigate the generalizability and robustness to sampling of the SOTA causal discovery approaches on additional simulated datasets with known ground truth. Our results reveal the limitations of existing causal modeling approaches when applied to large-scale, noisy, high-dimensional data with unobserved variables and unknown relationships between them. We show that the SOTA causal models explored in our experiments are not designed to take advantage from vasts amounts of data and have difficulty recovering ground truth when latent confounders are present; they do not generalize well across simulation scenarios and are not robust to sampling; they are vulnerable to data and modeling assumptions, and therefore, the results are hard to reproduce. Finally, when we outline lessons learned and provide recommendations to improve models for causal discovery and prediction of human social behavior from observational data, we highlight the importance of learning data to knowledge representations or transformations to improve causal discovery and describe the benefit of causal feature selection for predictive and prescriptive modeling.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2024

Boosting Barlow Twins Reduced Order Modeling for Machine Learning‐Based Surrogate Models in Multiphase Flow Problems

Abstract We present an innovative approach called boosting Barlow Twins reduced order modeling (BBT‐ROM) to enhance the reliability of machine learning surrogate models for multiphase flow problems. BBT‐ROM builds upon Barlow Twins reduced order modeling that leverages self‐supervised learning to effectively handle linear and nonlinear manifolds by constructing well‐structured latent spaces of input parameters and output quantities. To address the challenge of high contrast data in multiphase flow problems due to injection wells and faults, we employ a boosting algorithm within BBT‐ROM. This algorithm sequentially trains a set of weak models (i.e., inaccurate models), improving prediction accuracy through ensemble learning. To evaluate the performance of BBT‐ROM, we conduct three three‐dimensional multiphase flow problems, including waterflooding and geologic carbon storage (GCS), with varying numbers of input parameter cases and model domain features. The results demonstrate that BBT‐ROM excels at predicting non‐wetting phase saturation (e.g., oil or saturation) and fluid pressure, with average relative errors ranging from 0.5% to 3%. Importantly, BBT‐ROM showcases robustness when faced with limited input parameter space during GCS testing.

58 GEOSCIENCES↗

DOE OSTI2023

TransformerG2G: Adaptive time-stepping for learning temporal graph embeddings using transformers

Dynamic graph embedding has emerged as a very effective technique for addressing diverse temporal graph analytic tasks (i.e., link prediction, node classification, recommender systems, anomaly detection, and graph generation) in various applications. Such temporal graphs exhibit heterogeneous transient dynamics, varying time intervals, and highly evolving node features throughout their evolution. Hence, incorporating long-range dependencies from the historical graph context plays a crucial role in accurately learning their temporal dynamics. In this paper, we develop a graph embedding model with uncertainty quantification, TransformerG2G, by exploiting the advanced transformer encoder to first learn intermediate node representations from its current state (t) and previous context (over timestamps [t–1,t–l], l is the length of context). Moreover, we employ two projection layers to generate lower-dimensional multivariate Gaussian distributions as each node's latent embedding at timestamp t. We consider diverse benchmarks with varying levels of "novelty" as measured by the TEA (Temporal Edge Appearance) plots. Here, our experiments demonstrate that the proposed TransformerG2G model outperforms conventional multi-step methods and our prior work (DynG2G) in terms of both link prediction accuracy and computational efficiency, especially for high degree of novelty. Furthermore, the learned time-dependent attention weights across multiple graph snapshots reveal the development of an automatic adaptive time stepping enabled by the transformer. Importantly, by examining the attention weights, we can uncover temporal dependencies, identify influential elements, and gain insights into the complex interactions within the graph structure. For example, we identified a strong correlation between attention weights and node degree at the various stages of the graph topology evolution.

97 MATHEMATICS AND COMPUTING↗