Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “vision transformers”

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
DOE OSTI2024

Provable Repair of Vision Transformers

Vision Transformers have emerged as state-of-the-art image recognition tools, but may still exhibit incorrect behavior. Incorrect image recognition can have disastrous consequences in safety-critical real-world applications such as self-driving automobiles. In this paper, we present Provable Repair of Vision Transformers (PRoViT), a provable repair approach that guarantees the correct classification of images in a repair set for a given Vision Transformer without modifying its architecture. PRoViT avoids negatively affecting correctly classified images (drawdown) by minimizing the changes made to the Vision Transformer’s parameters and original output. Here, we observe that for Vision Transformers, unlike for other architectures such as ResNet or VGG, editing just the parameters in the last layer achieves correctness guarantees and very low drawdown. We introduce a novel method for editing these last-layer parameters that enables PRoViT to efficiently repair state-of-the-art Vision Transformers for thousands of images, far exceeding the capabilities of prior provable repair approaches.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Sequence length scaling in vision transformers for scientific images on frontier

Vision Transformers (ViTs) are pivotal for foundational models in scientific imagery, including Earth science applications, due to their capability to process large sequence lengths. While transformers for text have inspired scaling sequence lengths in ViTs, adapting these for ViTs introduces unique challenges. We develop distributed sequence parallelism for ViTs, enabling them to handle up to 1M tokens. Our approach, leveraging DeepSpeed-Ulysses and Long-Sequence-Segmentation with model sharding, is the first to apply sequence parallelism in ViT training, achieving a 94% batch scaling efficiency on 2,048 AMD-MI250X GPUs. Evaluating sequence parallelism in ViTs, particularly in models up to 10B parameters, highlighted substantial bottlenecks. We countered these with hybrid sequence, pipeline, and flash attention strategies, to scale beyond single GPU memory limits. Our method significantly enhances climate modeling accuracy by 20% in temperature predictions, marking the first training of a vision transformer model to convergence with a sequence length of 188K tokens, using full self-attention.

Tsaris, Aristeidis (aris) [ORNL] (ORCID:0000000277↗
DOE OSTI2024

Hyperparameter Studies for Vision Transformers Trained on High-Fidelity Simulations

This library is a collection of python modules that define, train, and analyze vision-transformer (ViT) machine learning models. The code implements, with mild modifications, ViT models that have been made publicly available through publication and GitHub code. The training data for these models is hydrodynamic simulation output in the form of numpy arrays. This library contains code to train these ViT models on the hydrodynamic simulation output with a variety of hyperparameters, and to compare the results of such models. Furthermore, the library contains definitions of simple convolutional neural network (CNN) machine learning architectures which can be trained on the same hydrodynamic simulation output. These are included as a reference point to compare the ViT models to. Additionally, the library includes trained ViT and CNN models and example input data for demonstration purposes. The code is based on the PyTorch python library.

Callis, Skylar↗
DOE OSTI2025

Permeability Prediction Using Vision Transformers

Accurate permeability predictions remain pivotal for understanding fluid flow in porous media, influencing crucial operations across petroleum engineering, hydrogeology, and related fields. Traditional approaches, while robust, often grapple with the inherent heterogeneity of reservoir rocks. With the advent of deep learning, convolutional neural networks (CNNs) have emerged as potent tools in image-based permeability estimation, capitalizing on micro-CT scans and digital rock imagery. This paper introduces a novel paradigm, employing vision transformers (ViTs)—a recent advancement in computer vision—for this crucial task. ViTs, which segment images into fixed-sized patches and process them through transformer architectures, present a promising alternative to CNNs. We present a methodology for implementing ViTs for permeability prediction, its results on diverse rock samples, and a comparison against conventional CNNs. The prediction results suggest that, with adequate training data, ViTs can match or surpass the predictive accuracy of CNNs, especially in rocks exhibiting significant heterogeneity. This study underscores the potential of ViTs as an innovative tool in permeability prediction, paving the way for further research and integration into mainstream reservoir characterization workflows.

58 GEOSCIENCES↗
DOE OSTI2025

SHF: Symmetrical Hierarchical Forest with Pretrained Vision Transformer Encoder for High-Resolution Medical Segmentation

This paper presents a novel approach to addressing the long-sequence problem in high-resolution medical images for Vision Transformers (ViTs). Using smaller patches as tokens can enhance ViT performance, but quadratically increases computation and memory requirements. Therefore, the common practice for applying ViTs to high-resolution images is either to: (a) employ complex sub-quadratic attention schemes or (b) use large to medium-sized patches and rely on additional mechanisms within the model to capture the spatial hierarchy of details. We propose Symmetrical Hierarchical Forest (SHF), a lightweight approach that adaptively patches the input image to increase token information density and encode hierarchical spatial structures into the input embedding. We then apply a reverse depatching scheme to the output embeddings of the transformer encoder, eliminating the need for convolution-based decoders. Unlike previous methods that modify attention mechanisms or use a complex hierarchy of interacting models, SHF can be retrofitted to any ViT model to allow it to learn the hierarchical structure of details in high-resolution images without requiring architectural changes. Experimental results demonstrate significant gains in computational efficiency and performance: on the PAIP WSI dataset, we achieved a 3∼32×speedup or a 2.95%∼7.03% increase in accuracy (measured by Dice score) at a 64K2 resolution with the same computational budget, compared to state-of-the-art production models. On the 3D medical datasets BTCV and KiTS, training was 6×faster, with accuracy gains of 6.93% and 5.9%, respectively, compared to models without SHF.

Zhang, Enzhi [Hokkaido University, Japan]↗
DOE OSTI2025

Compressing Vision Transformers in Geospatial Transfer Learning with Manifold-Constrained Optimization

Deploying geospatial foundation models on resource-constrained edge devices demands compact architectures that maintain high downstream performance. However, their large parameter counts and the accuracy loss often induced by compression limit practical adoption.In this work, we leverage manifold-constrained optimization framework DLRT to compress large vision transformer–based geospatial foundation models during transfer learning. By enforcing structured low-dimensional parameterizations aligned with downstream objectives, this approach achieves strong compression while preserving task-specific accuracy. We show that the method outperforms of-the-shelf low-rank methods as LoRA. Experiments on diverse geospatial benchmarks confirm substantial parameter reduction with minimal accuracy loss, enabling high-performing, on-device geospatial models.

Snyder, Thomas [Yale University]↗
DOE OSTI2024

Improving streamflow predictions across CONUS by integrating advanced machine learning models and diverse data

Accurate streamflow prediction is crucial to understand climate impacts on water resources and develop effective adaption strategies. A global long short-term memory (LSTM) model, using data from multiple basins, can enhance streamflow prediction, yet acquiring detailed basin attributes remains a challenge. To overcome this, we introduce the Geo-vision transformer (ViT)-LSTM model, a novel approach that enriches LSTM predictions by integrating basin attributes derived from remote sensing with a ViT architecture. Applied to 531 basins across the Contiguous United States, our method demonstrated superior prediction accuracy in both temporal and spatiotemporal extrapolation scenarios. Geo-ViT-LSTM marks a significant advancement in land surface modeling, providing a more comprehensive and effective tool for better understanding the environment responses to climate change.

Tayal, Kshitij↗
DOE OSTI2025

Visual Instance-aware Prompt Tuning

Visual Prompt Tuning (VPT) has emerged as a parameter-efficient fine-tuning paradigm for vision transformers, with conventional approaches utilizing dataset-level prompts that remain the same across all input instances. We observe that this strategy results in sub-optimal performance due to high variance in downstream datasets. To address this challenge, we propose Visual Instance-aware Prompt Tuning (ViaPT), which generates instance-aware prompts based on each individual input and fuses them with dataset-level prompts, leveraging Principal Component Analysis (PCA) to retain important prompting information. Moreover, we reveal that VPT-Deep and VPT-Shallow represent two corner cases based on a conceptual understanding, in which they fail to effectively capture instance-specific information, while random dimension reduction on prompts only yields performance between the two extremes. Instead, ViaPT overcomes these limitations by balancing dataset-level and instance-level knowledge, while reducing the amount of learnable parameters compared to VPT-Deep. Extensive experiments across 34 diverse datasets demonstrate that our method consistently outperforms state-of-the-art baselines, establishing a new paradigm for analyzing and optimizing visual prompts for vision transformers.

Xiao, Xi [ORNL] (ORCID:0009000009316982)↗
DOE OSTI2026

MATEY: multiscale adaptive transformer models for spatiotemporal physical systems

Accurate representation of the multiscale features in spatiotemporal physical systems using vision transformer architectures requires extremely long, computationally prohibitive token sequences. To address this issue, we propose two novel adaptive tokenization schemes that dynamically adjust patch sizes based on local features: one ensures convergent behavior to uniform patch refinement, while the other offers better computational efficiency. Moreover, we present a set of spatiotemporal attention schemes, where the temporal or axial spatial dimensions are decoupled, to evaluate their baseline computational and data efficiencies and to determine whether adaptive tokenization can improve this performance. We assess the performance of the proposed multiscale adaptive model, MATEY, in a sequence of experiments. Compared to a full spatiotemporal attention scheme or a scheme that decouples only the temporal dimension, we find that fully decoupled axial attention is less efficient and expressive, requiring more training time and model parameters to achieve the same accuracy. The experiments on the adaptive tokenization schemes show that, compared to a uniformly refined model, the proposed schemes achieve comparable or improved accuracy at a much lower cost in the tested two-dimensional settings. While the asymptotic analysis suggests the potential for favorable scaling, empirical validation at substantially longer sequence lengths remains to be performed in future work. Finally, we demonstrate in two fine-tuning tasks featuring different physics that models pretrained on PDEBench data outperform the ones trained from scratch, especially in the low data regime with frozen attention.

adaptive tokenization↗
DOE OSTI2024

ThermoPore: Predicting part porosity based on thermal images using deep learning

Part qualification is often a critical and labor-intensive process in additive manufacturing, particularly in the detection of defects such as porosity, which stands to benefit significantly from advancements in machine learning. We present a deep learning approach for quantifying and localizing ex-situ porosity within Laser Powder Bed Fusion fabricated samples utilizing in-situ thermal image monitoring data. Our goal is to build the real time porosity map of parts based on thermal images acquired during the build. The quantification task builds upon the established Convolutional Neural Network model architecture to predict pore count and the localization task leverages the spatial and temporal attention mechanisms of the novel Video Vision Transformer model to indicate areas of expected porosity. Our model for porosity quantification achieved a R 2 score of 0.57 and our model for porosity localization produced an average Intersection over Union (IoU) score of 0.32 and a maximum of 1.0. This work is setting the foundations of part porosity “Digital Twins” based on additive manufacturing monitoring data and can be applied downstream to reduce time-intensive post-inspection and testing activities during part qualification and certification. In addition, we seek to accelerate the acquisition of crucial insights normally only available through ex-situ part evaluation by means of machine learning analysis of in-situ process monitoring data.

Deep learning↗
DOE OSTI2024

Forecasting high-dimensional spatio-temporal systems from sparse measurements

This paper introduces a new neural network architecture designed to forecast high-dimensional spatio-temporal data using only sparse measurements. The architecture uses a two-stage end-to-end framework that combines neural ordinary differential equations (NODEs) with vision transformers. Initially, our approach models the underlying dynamics of complex systems within a low-dimensional space; and then it reconstructs the corresponding high-dimensional spatial fields. Many traditional methods involve decoding high-dimensional spatial fields before modeling the dynamics, while some other methods use an encoder to transition from high-dimensional observations to a latent space for dynamic modeling. In contrast, our approach directly uses sparse measurements to model the dynamics, bypassing the need for an encoder. This direct approach simplifies the modeling process, reduces computational complexity, and enhances the efficiency and scalability of the method for large datasets. We demonstrate the effectiveness of our framework through applications to various spatio-temporal systems, including fluid flows and global weather patterns. Although sparse measurements have limitations, our experiments reveal that they are sufficient to forecast system dynamics accurately over long time horizons. Our results also indicate that the performance of our proposed method remains robust across different sensor placement strategies, with further improvements as the number of sensors increases. This robustness underscores the flexibility of our architecture, particularly in real-world scenarios where sensor data is often sparse and unevenly distributed.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Transformer Masked Autoencoders for RF Device Fingerprinting

Machine learning methods for RF device fingerprinting typically rely on CNN-based models. Transformer-based models have outperformed CNNs for modulation classification tasks, but there are few implementations for device fingerprinting. We train a transformer for device fingerprinting with the largest device count to date and explore several variations of the architecture. Additionally, we demonstrate that pre-training an RF transformer as a Masked Autoencoder improves classification accuracy, as has been observed for CNN fingerprinting models and vision transformers.

artificial intelligence↗
DOE OSTI2025

Micrometer: Micromechanics transformer for predicting full field mechanical responses of heterogeneous materials

Predicting mechanical responses of heterogeneous materials across scales remains a significant challenge. Traditional computational methods often struggle with complex and multiscale nature of these materials, limiting their effectiveness in real-world applications. Here, in this paper, we introduce Micrometer, a vision transformer based deep learning model designed to predict full field mechanical responses of heterogeneous materials, bridging the gap between computer vision and solid mechanics problems. We show that Micrometer, trained on a large-scale high-resolution dataset of 2D fiber-reinforced composites, can achieve state-of-the-art performance in predicting microscale strain fields across a wide range of material properties and loading conditions. Our model demonstrates accuracy and computational efficiency in applications such as computational homogenization and multiscale modeling, reducing computational time by up to two orders of magnitude compared to conventional numerical solvers while maintaining less than 1 % errors in predicting macroscale stress fields. Furthermore, we showcase Micrometer’s adaptability through transfer learning experiments on new materials with limited data, highlighting its potential to tackle diverse scenarios in computational solid mechanics. These results represent a significant step towards AI-driven innovation in materials science, addressing the limitations of traditional numerical methods and paving the way for more efficient simulations of heterogeneous materials across various industrial applications.

Composite materials↗
DOE OSTI2024

Transformer Neural Networks with Spatiotemporal Attention for Predictive Control and Optimization of Industrial Processes

In the context of real-time optimization and model predictive control of industrial systems, machine learning, and neural networks represent cutting-edge tools that hold promise for enhancing dynamic modeling. This work presents a novel transformer neural network architecture for real-time optimization and model predictive control. This network design includes a modified attention mechanism inspired by positional embedding attention from vision transformers and task-specific modifications to the input-output structure of the transformer’s decoder stack. Experiments were conducted using data from a 450 MW coal-fired power plant to evaluate this approach's effectiveness. The transformer neural network was compared with conventional recurrent models, including GRU and LSTM. The transformer exhibited a 6% increase in the R-squared (R2) value of predictions and an 83% reduction in mean squared error (MSE). Computation time was also reduced by 84% compared to conventional recurrent models.

Gallup, Ethan R.↗