Search NASA⌕ Search

SEARCH · Search NASA

Results for “adversarial attack”

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 OSTI2025

Deploying Adversarial Attacks in Super-Resolution Models

Reliable super-resolution methods are crucial for applications like remote sensing, grid resilience and disaster impact analysis, and standoff biometrics. These methods infuse additional high-frequency information into reconstructions, allowing for better contextualization and image intelligence. However, super-resolution models can also introduce hallucinations or other unseen vulnerabilities that could be exploited by an adversary. This is further compounded by the prominence of deep learning in these models, as models are often blindly applied on out-of-distribution images. In this work, we implement adversarial attacks in common open-source super-resolution models and examine their impact on reconstructions and downstream classification tasks. We find that an adversarially trained super-resolution model can produce high-quality reconstructions that degrade downstream classifications. Moreover, these attacks do not require access to low-resolution imagery or class labels at inference time. These results demonstrate the vulnerability of super-resolution methods to malicious actors and motivates the development of a detector for super-resolution adversarial attacks. Further exploration of adversarial attacks in this domain is required to ensure trustworthiness and robustness of super-resolution models for national security applications.

45 MILITARY TECHNOLOGY, WEAPONRY, AND NATIONAL DEF↗
DOE OSTI2025

Dynamical Low-Rank Compression of Neural Networks with Robustness under Adversarial Attacks

Deployment of neural networks on resource-constrained devices demands models that are both compact and robust to adversarial inputs. However, compression and adversarial robustness often conflict. In this work, we introduce a dynamical low-rank training scheme enhanced with a novel spectral regularizer that controls the condition number of the low-rank core in each layer. This approach mitigates the sensitivity of compressed models to adversarial perturbations without sacrificing clean accuracy. The method is model- and data-agnostic, computationally efficient, and supports rank adaptivity to automatically compress the network at hand. Extensive experiments across standard architectures, datasets, and adversarial attacks show the regularized networks can achieve over 94 compression while recovering or improving adversarial accuracy relative to uncompressed baselines.

Schotthoefer, Steffen [ORNL] (ORCID:00000002156965↗
DOE OSTI2025

XSub: Explanation-Driven Adversarial Attack against Blackbox Classifiers via Feature Substitution

Despite its significant benefits in enhancing the transparency and trustworthiness of artificial intelligence (AI) systems, explainable AI (XAI) can unintentionally provide adversaries with insights into blackbox models, increasing their vulnerability to various attacks. In this paper, we develop a novel explanation-driven adversarial attack against blackbox classifiers based on feature substitution, called XSub. The key idea of XSub is to strategically replace important features (identified via XAI) in the original sample with corresponding important features of a different label, thereby increasing the likelihood of the model misclassifying the perturbed sample. XSub only requires a minimal number of queries and can be easily extended to launch backdoor attacks in case the attacker has access to the model's training data. Our evaluation shows that XSub is not only effective and stealthy but also low-cost, showcasing its feasibility across a wide range of AI applications.

adversarial attack↗
DOE OSTI2025

Persistent Classification: Understanding Adversarial Attacks by Studying Decision Boundary Dynamics

ABSTRACT There are a number of hypotheses underlying the existence of adversarial examples for classification problems. These include the high‐dimensionality of the data, the high codimension in the ambient space of the data manifolds of interest, and that the structure of machine learning models may encourage classifiers to develop decision boundaries close to data points. This article proposes a new framework for studying adversarial examples that does not depend directly on the distance to the decision boundary. Similarly to the smoothed classifier literature, we define a (natural or adversarial) data point to be ( γ , σ)‐stable if the probability of the same classification is at least for points sampled in a Gaussian neighborhood of the point with a given standard deviation . We focus on studying the differences between persistence metrics along interpolants of natural and adversarial points. We show that adversarial examples have significantly lower persistence than natural examples for large neural networks in the context of the MNIST and ImageNet datasets. We connect this lack of persistence with decision boundary geometry by measuring angles of interpolants with respect to decision boundaries. Finally, we connect this approach with robustness by developing a manifold alignment gradient metric and demonstrating the increase in robustness that can be achieved when training with the addition of this metric.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2025

Exploring Black-box Adversarial Attacks on Low-rank Constrained Neural Networks

Low-rank compression has been shown as an effective tool to reduce parameter counts of convolutional and vision transformer architectures; however, low-rank training often reduces model robustness to adversarial perturbations. In this work, we explore the effects of low-rank training on black-box attacks, where attacked images are generated without knowledge of the low-rank parameters. We find that low-rank training is not sufficient as a black-box defense and can sometimes produce worse than expected as compared to baseline models. Influencing the spectrum of the low-rank models during training, which is known to increase model robustness against white-box attacks, improves black-box performance as well.

Schnake, Stefan [ORNL] (ORCID:0000000215183538)↗
DOE OSTI2024

Semantic Stealth: Crafting Covert Adversarial Patches for Sentiment Classifiers Using Large Language Models

Deep learning models have been shown to be vulnerable to adversarial attacks, in which perturbations to their inputs cause the model to produce incorrect predictions. As opposed to adversarial attacks in computer vision, where small changes introduced to pixel values can drastically alter a model's output while remaining imperceptible to humans, text-based attacks are difficult to conceal due to the discrete nature of tokens. Consequently, unconstrained gradient-based attacks often produce adversarial examples that lack semantic meaning, rendering them detectable through visual inspection or perplexity filters. In contrast to methods that rely on gradient-based optimization in the embedding space, we propose an approach that leverages a Large Language Model's ability to generate grammatically correct and semantically meaningful text to craft adversarial patches that seamlessly blend in with the original input text. These patches can be used to alter the behavior of a target model, such as a text classifier. Since our approach does not rely on gradient backpropagation, it only requires access to the target model's confidence scores, making it a grey-box attack. We demonstrate the feasibility of our approach using open-source LLMs, including Intel's Neural Chat, Llama2, and Mistral-Instruct, to generate adversarial patches capable of altering the predictions of a distilBERT model fine-tuned on the IMDB reviews dataset for sentiment classification.

Roa Carvajal, Maria↗
DOE OSTI2026

Community detection robustness of graph neural networks

Graph neural networks (GNNs) are increasingly widely used for community detection in attributed networks. They combine structural topology with node attributes through message passing and pooling. However, their robustness or lack thereof with respect to different perturbations and targeted attacks in conjunction with community detection tasks is not well understood. To shed light on latent mechanisms behind GNN sensitivity on community detection tasks, we conduct a systematic computational evaluation of six widely adopted GNN architectures graph convolutional network, graph attention network, graph sample and aggregate (GraphSAGE), differentiable pooling (DiffPool), minimum cut pooling (MinCUT), and deep modularity networks (DMoN). The analysis covers three perturbation categories: node attribute manipulations, edge topology distortions, and adversarial attacks. We use element-centric similarity as the evaluation metric on synthetic benchmarks and real-world citation networks. Our findings indicate that supervised GNNs tend to achieve higher baseline accuracy, while unsupervised methods, particularly DMoN, maintain stronger resilience under targeted and adversarial perturbations. Furthermore, robustness appears to be strongly influenced by community strength, with well-defined communities reducing performance loss. Across all models, node attribute perturbations associated with targeted edge deletions and shifts in attribute distributions tend to cause the largest degradation in community recovery. These findings highlight important trade-offs between accuracy and robustness in GNN-based community detection and offer insights into selecting architectures resilient to noise and adversarial attacks.

Goel, Jaidev [Virginia Polytechnic Inst. and State↗
DOE OSTI2024

Hiding-in-Plain-Sight (HiPS) Attack on CLIP for Targetted Object Removal from Images

Machine learning models are known to be vulnerable to adversarial attacks, but prior works have mostly focused on single-modalities. With the rise of large multi-modal models (LMMs) like CLIP, which combine vision and language capabilities, new vulnerabilities have emerged. However, these multimodal targeted attacks aim to completely change the model's output to what the adversary wants. In many realistic scenarios, an adversary might seek to make only subtle modifications to the output, so that the changes go unnoticed by downstream models or even by humans. We introduce Hiding-in-Plain-Sight (HiPS) attacks, a novel class of adversarial attacks that subtly modifies model predictions by selectively concealing target object(s), as if the target object was absent from the scene. We propose two HiPS attack variants, HiPS-cls and HiPS-cap, and demonstrate their effectiveness in transferring to downstream image captioning models, such as CLIP-Cap, for targeted object removal from image captions.

Daw, Arka [ORNL] (ORCID:0009000633191271)↗
DOE OSTI2025

Performance Evaluation of Vertical Federated Machine Learning Against Adversarial Threats on Wide-Area Control System: Preprint

Federated machine learning (FL) is gaining significant popularity to develop cybersecurity solutions in power grids because of its advanced capability to support decentralized data handing at local devices, its privacy preservation, and its low-bandwidth requirement. However, the evolving adversarial machine learning (AML) threats raise significant concerns for the cybersecurity of FL architectures. The FL-based split neural network (SplitNN) achieves high performance through the decentralized training of local neural network models while preserving data privacy across multiple entities. In this paper, we propose a methodology for evaluating the performance of a vertical FLbased anomaly detector against different types of AML attacks, including denial-of-service attacks, adversarial data injection attacks, and replay attacks on the trained local models deployed in the grid network. For a case study, we consider the modified IEEE 13-bus system, and we develop SplitNN-based binary and multiclass classification models to detect, locate, and identify different types of data integrity attacks on the volt-watt control with two pooling layers: maximum pooling and AvgPool. Our experimental results, computed through performance metrics, reveal that the severity of these AML attacks varies with the integrated pooling mechanism, the type of classification model, and the nature of the cyberattack. Further, the AML attacks negatively impacted the prediction time per sample for the pretrained SplitNN during the online testing.

adversarial threats↗
DOE OSTI2024

Universal Fourier Attack for Time Series

A wide variety of adversarial attacks have been proposed and explored using image and audio data. These attacks are notoriously easy to generate digitally when the attacker can directly manipulate the input to a model, but are much more difficult to implement in the real world. In this paper we present a universal, time invariant attack for general time series data such that the attack has a frequency spectrum primarily composed of the frequencies present in the original data. The universality of the attack makes it fast and easy to implement as no computation is required to add it to an input, while time invariance is useful for real world deployment. Additionally, the frequency constraint ensures the attack can withstand filtering defenses. We demonstrate the effectiveness of the attack on two different classification tasks through both digital and real world experiments, and show that the attack is robust against common transform-and-compare defense pipelines.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Microreactor Security-by-Design Recommendations for Domestic and International Deployments

This report outlines methods vendors can use to incorporate security-by-design (SeBD) into their microreactor facility design to support and address security for both U.S. and international deployment. The team developed a hypothetical below-grade microreactor with a physical protection system (PPS) to protect the microreactor against acts of theft and sabotage and evaluated it against two adversary attack scenarios defined by a group of adversary subject matter experts (SMEs). The hypothetical microreactor facility consists of two distinct buildings. The first is the above-grade protected area (PA) entry control point (ECP) building, which houses security personnel responsible for conducting screenings and managing access to the PA. The second building is the reactor building, which features both an above-grade floor and a below-grade floor.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2026

Quantum-inspired weight-constrained neural network: Reducing variable numbers by 100× compared to standard neural networks

Although quantum machine learning has shown great promise, the practical application of quantum computers remains constrained in the noisy intermediate-scale quantum era. To take advantage of quantum machine learning, we investigate the underlying mathematical principles of these quantum models and find that the quantum neural network with amplitude encoding is equivalent to a weight-constrained neural network. Motivated by this discovery, we develop a classical weight-constrained neural network. We find that this approach can reduce the number of variables in a classical neural network by a factor of 135 while preserving its accuracy. In addition, we develop a dropout method to enhance the robustness of quantum machine learning models, which are highly susceptible to adversarial attacks. This technique can also be applied to improve the adversarial robustness of the classical weight-constrained neural network, which is essential for industry applications, such as self-driving vehicles. Our work offers an approach to reduce the complexity of large classical neural networks, addressing a critical challenge in machine learning.

quantum algorithms & computation↗
DOE OSTI2025

Dynamic Low-Rank Training with Spectral Regularization: Achieving Robustness in Compressed Representations

Deployment of neural networks on resource-constrained devices demands models that are both compact and robust to adversarial inputs. However, compression and adversarial robustness often conflict. In this work, we introduce a dynamical low-rank training scheme enhanced with a novel spectral regularizer that controls the condition number of the low-rank core in each layer. This approach mitigates the sensitivity of compressed models to adversarial perturbations without sacrificing clean accuracy. The method is model- and data-agnostic, computationally efficient, and supports rank adaptivity to automatically compress the network at hand. Extensive experiments across standard architectures, datasets, and adversarial attacks show the regularized networks can achieve over 94\% compression while recovering or improving adversarial accuracy relative to uncompressed baselines.

Schotthoefer, Steffen [ORNL] (ORCID:00000002156965↗
DOE OSTI2024

A Compound Data Poisoning Technique with Significant Adversarial Effects on Transformer-based Sentiment Classification Tasks

Transformer-based models have demonstrated much success in various natural language processing tasks. However, they are often vulnerable to adversarial attacks, such as data poisoning, which can intentionally fool the model into generating incorrect results. In this article, we present a novel, compound variant of a data poisoning attack on a transformer-based model that maximizes the poisoning effect while minimizing the scope of poisoning. Here we do so by combining the established data poisoning technique (label flipping) with a novel adversarial artifact selection and insertion technique aimed at minimizing detectability and the scope of the poisoning footprint. We find that by using a combination of these two techniques, we achieve a state-of-the-art attack success rate of approximately 90% while poisoning only 0.5% of the original training set, thus minimizing the scope and detectability of the poisoning action. These findings have the potential to advance the development of better data poisoning detection methods.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Emerging Threats in Transportation Security Related to Intelligent Transportation Systems (ITS)

Transport of high-consequence shipments requires a resilient and robust systems of systems to guarantee cargo arrival. Furthermore, rising adoption of technologies such as connected and automated vehicles (CAVs), intelligent infrastructure, and vehicle-to-everything (V2X) communication presents unique challenges for securing transportation systems. Within these Intelligent Transportation Systems (ITS), several additional vulnerabilities exist that create pathways for adversarial attacks and cargo interception. For example, connectivity provides cyber pathways directly into vehicle systems and infrastructure for malicious actors. Furthermore, advanced vehicle automation exposes additional vehicle control necessary for shipment interception otherwise unavailable to adversaries. Within this paper, we will discuss the specific threats introduced by ITS-enabled technologies currently deployed and in development. These include those mentioned related to connectivity and automation, but will be expanded into grid, infrastructure, and vehicle specific threats. In addition, we will discuss how to potentially mitigate these emerging challenges as well as how to safeguard transportation systems from next generation attacks.

Cook, Adian [ORNL] (ORCID:0000000160825395)↗
DOE OSTI2025

Robustness of Deep Learning Classification to Adversarial Input on GPUs: Asynchronous Parallel Accumulation Is a Source of Vulnerability

The ability of machine learning (ML) classification models to resist small, targeted input perturbations—known as adversarial attacks—is a key measure of their safety and reliability. We show that floating-point non associativity (FPNA) coupled with asynchronous parallel programming on GPUs is sufficient to result in misclassification, without any perturbation to the input. Additionally, we show that this misclassification is particularly significant for inputs close to the decision boundary and that standard adversarial robustness results may be overestimated up to 4.6 when not considering machine-level details. We first study a linear classifier, before focusing on standard Graph Neural Network (GNN) architectures and datasets used in robustness assessments. We develop a novel black-box attack using Bayesian optimization to discover external workloads that can change the instruction scheduling which bias the output of reductions on GPUs and reliably lead to misclassification. Motivated by these results, we present a new learnable permutation (LP) gradient-based approach to learning floating-point operation orderings that lead to misclassifications. The LP approach provides a worst-case estimate in a computationally efficient manner, avoiding the need to run identical experiments tens of thousands of times over a potentially large set of possible GPU states or architectures. Finally, using instrumentation-based testing, we investigate parallel reduction ordering across different GPU architectures under external background workloads, when utilizing multi-GPU virtualization, and when applying power capping. Our results demonstrate that parallel reduction ordering varies significantly across architectures under the first two conditions, substantially increasing the search space required to fully test the effects of this parallel scheduler-based vulnerability. These results and the methods developed here can help to include machine-level considerations into adversarial robustness assessments, which can make a difference in safety and mission critical applications.

Shanmugavelu, Sanjif [Maxeler Technologies, a Groq↗
DOE OSTI2025

Quantum adversarial learning for kernel methods

We show that hybrid quantum classifiers based on quantum kernel methods and support vector machines are vulnerable against adversarial attacks, namely small engineered perturbations of the input data can deceive the classifier into predicting the wrong result. Nonetheless, we also show that simple defense strategies based on data augmentation with a few crafted perturbations can make the classifier robust against new attacks. Our results find applications in security-critical learning problems and in mitigating the effect of some forms of quantum noise, since the attacker can also be understood as part of the surrounding environment.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗