Results for “natural language processing” — Page 6

DOE OSTI2026

Hybrid Quantum–Classical Graph Transformers for Efficient Sentiment Analysis

Quantum Machine Learning (QML) offers a promising paradigm that leverages quantum computing principles to develop efficient and expressive models for learning from complex and structured data. Recent advances in natural language processing (NLP) and artificial intelligence (AI) have demonstrated capabilities in understanding, generating, and reasoning over linguistic and multimodal information. In this work, we present the Quantum Graph Transformer (QGT), a hybrid quantum–classical architecture that extends graph transformer capabilities through quantum self-attention. The QGT models variable-length sentences as token graphs, where both the embedding encoding and the self-attention mechanisms are implemented using parameterized quantum circuits (PQCs), enabling efficient contextual learning with significantly fewer trainable parameters. We train QGT using both fully connected and 𝑘 -nearest-neighbor graph structures and evaluate it on five benchmark sentiment-classification datasets. Experimental results show that QGT consistently achieves higher or comparable accuracy to existing quantum NLP models and outperforms a Classical Graph Transformer (CGT) baseline with identical architecture, achieving 29.4 × fewer parameters while requiring 3–5 × fewer samples to reach comparable performance. These findings highlight the potential of graph-based quantum models as scalable and data-efficient architectures for natural language understanding.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

NASA NTRSDate not supplied

Machine Learning Application in Aircraft Engine Conceptual Design

In the current competitive environment, the successful creation and application of machine learning (ML) technologies have become crucial across multiple industries. This study outlines the process of creating and implementing ML models for conceptualizing and evaluating aircraft engines. These models use supervised deep-learning algorithms to analyze patterns within an open-source repository containing data on both production and research conventional turbofan engines. Key focus areas include crucial engine parameters such as thrust-specific fuel consumption (TSFC), engine weight, engine diameter, and turbomachinery stage counts. While developing ML models is fundamental, ensuring their seamless deployment is equally important. To address this, a conversational AI chatbot is constructed using natural language processing (NLP) techniques to facilitate the deployment of these ML models. The comprehensive workflow includes several key stages: gathering and enhancing engine data, training and cross validating the ML models, testing and evaluating their performance, and finally, deploying, monitoring, and updating the ML models. By following this systematic approach, the aim is to streamline the development and deployment process of ML models tailored for aircraft engine conceptual design.

Aircraft Engine↗

NASA NTRS2008

From Informal Safety-Critical Requirements to Property-Driven Formal Validation

Most of the efforts in formal methods have historically been devoted to comparing a design against a set of requirements. The validation of the requirements themselves, however, has often been disregarded, and it can be considered a largely open problem, which poses several challenges. The first challenge is given by the fact that requirements are often written in natural language, and may thus contain a high degree of ambiguity. Despite the progresses in Natural Language Processing techniques, the task of understanding a set of requirements cannot be automatized, and must be carried out by domain experts, who are typically not familiar with formal languages. Furthermore, in order to retain a direct connection with the informal requirements, the formalization cannot follow standard model-based approaches. The second challenge lies in the formal validation of requirements. On one hand, it is not even clear which are the correctness criteria or the high-level properties that the requirements must fulfill. On the other hand, the expressivity of the language used in the formalization may go beyond the theoretical and/or practical capacity of state-of-the-art formal verification. In order to solve these issues, we propose a new methodology that comprises of a chain of steps, each supported by a specific tool. The main steps are the following. First, the informal requirements are split into basic fragments, which are classified into categories, and dependency and generalization relationships among them are identified. Second, the fragments are modeled using a visual language such as UML. The UML diagrams are both syntactically restricted (in order to guarantee a formal semantics), and enriched with a highly controlled natural language (to allow for modeling static and temporal constraints). Third, an automatic formal analysis phase iterates over the modeled requirements, by combining several, complementary techniques: checking consistency; verifying whether the requirements entail some desirable properties; verify whether the requirements are consistent with selected scenarios; diagnosing inconsistencies by identifying inconsistent cores; identifying vacuous requirements; constructing multiple explanations by enabling the fault-tree analysis related to particular fault models; verifying whether the specification is realizable.

Cimatti, Alessandro↗

NASA NTRSDate not supplied

An Approach to Identifying Aspects of Positive Pilot Behavior within the Aviation Safety Reporting System

The National Airspace System (NAS) is constantly evolving as air traffic continues to ramp up to pre-pandemic numbers and projected to grow to unprecedented levels in the coming years. As well as increasing demand to the current system, emerging operations such as Unmanned Autonomous Systems are also expected to add to complexity in the airspace. To address these issues, the industry and government agencies supporting the NAS will need to rely upon additional automation and new technologies to address future operational requirements, while continuing to be a world-leading safe transportation system. As these new technologies are implemented, the system continues to rely on human pilots and controllers in the loop to monitor the system and intervene in situations the automation cannot handle. The goal of proactively addressing safety is of foremost concern to ensure passenger confidence. The industry has implemented various Safety Monitoring Systems to identify safety risks and proactively address them before they result in a serious incident or accident. One such program is the Aviation Safety Reporting System (ASRS). ASRS is a long-established system where pilots and controllers voluntarily and anonymously report safety incidents they experienced and observed during line operations by providing rich text narratives describing the events, the environment, and conditions leading to the safety event of concern. These narratives provide insight and context around events of interest and can be used to identify emerging problems. They can trigger investigations within Flight Operational Quality Assurance or Flight Data Monitoring programs. However, this process typically focuses on the adverse events and the unsafe aspects of the operations surrounding the reported or detected events. This perspective of investigating factors that went wrong around an adverse event is commonly referred to as Safety I. Alternatively, characterizing successful actions that operators perform every day under varying conditions that keep the system within safe operating bounds is a concept referred to as Safety II. The benefit of the Safety II view is that the scope is much larger than that of Safety I since a vast majority of the operations result in successful flights. Many of the successful techniques used to manage operational threats are not documented in standard operating procedures or taught during training. They are typically acquired over time by working with experienced pilots during line operations or in many cases after experiencing a problem for the first time and reacting to it in situ, drawing from years of experience to manage the threat. In an attempt to quantify these positive actions, we are proposing an approach to extracting key behaviors within ASRS reports that can support the Safety II concept. Our analysis assumes that ASRS reports contain some descriptions of corrective actions that operators performed to prevent a situation from leading to an accident. Leveraging recent advances in Natural Language Process modeling, we have developed an approach to extract positive sentiment from reports, embed these positive statements in a vector space where they can be numerically analyzed, and clustering these statements into similar contextual categories. From these contextualized categories we can attempt to summarized and distilled aspects of the positive behavior. The goal is to identify categories of behavior that describe consistent operator techniques that supports the Safety II concept. With this information, airlines may enable learning from these positive actions, or address procedures that need to be changed to avoid having pilots implement a workaround. These insights can provide a lens into what is “going right” in the operations that may otherwise not be known widely within the community. It is envisioned that this approach can be extended to other narrative programs such as Line Operation Safety Audit or Learning Improvement Team reports where similar observed behavior can be analyzed to extract positive actions and inform the overall operations.

NLP↗

NASA NTRSDate not supplied

An Approach to Identifying Aspects of Positive Pilot Behavior within the Aviation Safety Reporting System

The National Airspace System (NAS) is constantly evolving as air traffic continues to ramp up to pre-pandemic numbers and projected to grow to unprecedented levels in the coming years. As well as increasing demand to the current system, emerging operations such as Unmanned Autonomous Systems are also expected to add to complexity in the airspace. To address these issues, the industry and government agencies supporting the NAS will need to rely upon additional automation and new technologies to address future operational requirements, while continuing to be a world-leading safe transportation system. As these new technologies are implemented, the system continues to rely on human pilots and controllers in the loop to monitor the system and intervene in situations the automation cannot handle. The goal of proactively addressing safety is of foremost concern to ensure passenger confidence. The industry has implemented various Safety Monitoring Systems to identify safety risks and proactively address them before they result in a serious incident or accident. One such program is the Aviation Safety Reporting System (ASRS). ASRS is a long-established system where pilots and controllers voluntarily and anonymously report safety incidents they experienced and observed during line operations by providing rich text narratives describing the events, the environment, and conditions leading to the safety event of concern. These narratives provide insight and context around events of interest and can be used to identify emerging problems. They can trigger investigations within Flight Operational Quality Assurance or Flight Data Monitoring programs. However, this process typically focuses on the adverse events and the unsafe aspects of the operations surrounding the reported or detected events. This perspective of investigating factors that went wrong around an adverse event is commonly referred to as Safety I. Alternatively, characterizing successful actions that operators perform every day under varying conditions that keep the system within safe operating bounds is a concept referred to as Safety II. The benefit of the Safety II view is that the scope is much larger than that of Safety I since a vast majority of the operations result in successful flights. Many of the successful techniques used to manage operational threats are not documented in standard operating procedures or taught during training. They are typically acquired over time by working with experienced pilots during line operations or in many cases after experiencing a problem for the first time and reacting to it in situ, drawing from years of experience to manage the threat. In an attempt to quantify these positive actions, we are proposing an approach to extracting key behaviors within ASRS reports that can support the Safety II concept. Our analysis assumes that ASRS reports contain some descriptions of corrective actions that operators performed to prevent a situation from leading to an accident. Leveraging recent advances in Natural Language Process modeling, we have developed an approach to extract positive sentiment from reports, embed these positive statements in a vector space where they can be numerically analyzed, and clustering these statements into similar contextual categories. From these contextualized categories we can attempt to summarized and distilled aspects of the positive behavior. The goal is to identify categories of behavior that describe consistent operator techniques that supports the Safety II concept. With this information, airlines may enable learning from these positive actions, or address procedures that need to be changed to avoid having pilots implement a workaround. These insights can provide a lens into what is “going right” in the operations that may otherwise not be known widely within the community. It is envisioned that this approach can be extended to other narrative programs such as Line Operation Safety Audit or Learning Improvement Team reports where similar observed behavior can be analyzed to extract positive actions and inform the overall operations.

NLP↗

NASA NTRSDate not supplied

An Approach to Identifying Aspects of Positive Pilot Behavior within the Aviation Safety Reporting System

The National Airspace System (NAS) is constantly evolving as air traffic continues to ramp up to pre-pandemic numbers and projected to grow to unprecedented levels in the coming years. As well as increasing demand to the current system, emerging operations such as Unmanned Autonomous Systems are also expected to add to complexity in the airspace. To address these issues, the industry and government agencies supporting the NAS will need to rely upon additional automation and new technologies to address future operational requirements, while continuing to be a world-leading safe transportation system. As these new technologies are implemented, the system continues to rely on human pilots and controllers in the loop to monitor the system and intervene in situations the automation cannot handle. The goal of proactively addressing safety is of foremost concern to ensure passenger confidence. The industry has implemented various Safety Monitoring Systems to identify safety risks and proactively address them before they result in a serious incident or accident. One such program is the Aviation Safety Reporting System (ASRS). ASRS is a long-established system where pilots and controllers voluntarily and anonymously report safety incidents they experienced and observed during line operations by providing rich text narratives describing the events, the environment, and conditions leading to the safety event of concern. These narratives provide insight and context around events of interest and can be used to identify emerging problems. They can trigger investigations within Flight Operational Quality Assurance or Flight Data Monitoring programs. However, this process typically focuses on the adverse events and the unsafe aspects of the operations surrounding the reported or detected events. This perspective of investigating factors that went wrong around an adverse event is commonly referred to as Safety I. Alternatively, characterizing successful actions that operators perform every day under varying conditions that keep the system within safe operating bounds is a concept referred to as Safety II. The benefit of the Safety II view is that the scope is much larger than that of Safety I since a vast majority of the operations result in successful flights. Many of the successful techniques used to manage operational threats are not documented in standard operating procedures or taught during training. They are typically acquired over time by working with experienced pilots during line operations or in many cases after experiencing a problem for the first time and reacting to it in situ, drawing from years of experience to manage the threat. In an attempt to quantify these positive actions, we are proposing an approach to extracting key behaviors within ASRS reports that can support the Safety II concept. Our analysis assumes that ASRS reports contain some descriptions of corrective actions that operators performed to prevent a situation from leading to an accident. Leveraging recent advances in Natural Language Process modeling, we have developed an approach to extract positive sentiment from reports, embed these positive statements in a vector space where they can be numerically analyzed, and clustering these statements into similar contextual categories. From these contextualized categories we can attempt to summarized and distilled aspects of the positive behavior. The goal is to identify categories of behavior that describe consistent operator techniques that supports the Safety II concept. With this information, airlines may enable learning from these positive actions, or address procedures that need to be changed to avoid having pilots implement a workaround. These insights can provide a lens into what is “going right” in the operations that may otherwise not be known widely within the community. It is envisioned that this approach can be extended to other narrative programs such as Line Operation Safety Audit or Learning Improvement Team reports where similar observed behavior can be analyzed to extract positive actions and inform the overall operations.

NLP↗

DOE OSTI2025

Towards Automated Assessment of Vulnerability Exposures in Security Operations

Current approaches for risk analysis of software vulnerabilities using manual assessment and numeric scoring do not complete fast enough to keep pace with the maintenance work rate to patch and mitigate the vulnerabilities. This paper proposes a new approach to modeling software vulnerability risk in the context of the network environment and firewall configuration. In the approach, vulnerability features are automatically matched up with networking, target asset, and adversary features to determine whether adversaries can exploit a vulnerability. The ability of adversaries to reach a vulnerability is modeled by automatically identifying the network services associated with vulnerabilities through a pipeline of machine learning and natural language processing and automatically analyzing network reachability. Our results show that the pipeline can identify network services accurately. We also find that only a small number of vulnerabilities pose real risks to a system. However, if left unmitigated, adversarial reach to vulnerabilities may extend to nullify the effect of firewall countermeasures.

Huff, Philip↗

DOE OSTI2026

Generalist multimodal AI: A review of architectures, challenges and opportunities

Multimodal models are expected to be a critical component to future advances in artificial intelligence. Here, this field is starting to grow rapidly with a surge of new design elements motivated by the success of foundation models in natural language processing (NLP) and vision. It is widely hoped that further extending the foundation models to multiple modalities (e.g., text, image, video, sensor, time series, graph, etc.) will ultimately lead to generalist multimodal models, i.e. one model across different data modalities and tasks. However, there is little research that systematically analyzes recent multimodal models (particularly the ones that work beyond text and vision) with respect to the underling architecture proposed. Therefore, this work provides a fresh perspective on generalist multimodal models (GMMs) via a novel architecture and training configuration specific taxonomy. This includes factors such as Unifiability, Modularity, and Adaptability that are pertinent and essential to the wide adoption and application of GMMs. The review further highlights key challenges and prospects for the field and guide the researchers into the new advancements.

Artificial intelligence (AI)↗

DOE OSTI2024

A Database of Stress-Strain Properties Auto-generated from the Scientific Literature using ChemDataExtractor

Abstract There has been an ongoing need for information-rich databases in the mechanical-engineering domain to aid in data-driven materials science. To address the lack of suitable property databases, this study employs the latest version of the chemistry-aware natural-language-processing (NLP) toolkit, ChemDataExtractor, to automatically curate a comprehensive materials database of key stress-strain properties. The database contains information about materials and their cognate properties: ultimate tensile strength, yield strength, fracture strength, Young’s modulus, and ductility values. 720,308 data records were extracted from the scientific literature and organized into machine-readable databases formats. The extracted data have an overall precision, recall and F-score of 82.03%, 92.13% and 86.79%, respectively. The resulting database has been made publicly available, aiming to facilitate data-driven research and accelerate advancements within the mechanical-engineering domain.

Kumar, Pankaj↗

DOE OSTI2025

Decoding substance use disorder severity from clinical notes using a large language model

Substance use disorder (SUD) poses a major concern due to its detrimental effects on health and society. SUD identification and treatment depend on a variety of factors such as severity, co-determinants (e.g., withdrawal symptoms), and social determinants of health. Existing diagnostic coding systems used by insurance providers, like the International Classification of Diseases (ICD-10), lack granularity for certain diagnoses, but American clinicians will add this granularity (as that found within the Diagnostic and Statistical Manual of Mental Disorders classification or DSM-5) as supplemental unstructured text in clinical notes. Traditional natural language processing (NLP) methods face limitations in accurately parsing such diverse clinical language. Large language models (LLMs) offer promise in overcoming these challenges by adapting to diverse language patterns. This study investigates the application of LLMs for extracting severity-related information for various SUD diagnoses from clinical notes. We propose a workflow employing zero-shot learning of LLMs with carefully crafted prompts and post-processing techniques. Through experimentation with Flan-T5, an open-source LLM, we demonstrate its superior recall compared to the rule-based approach. Focusing on 11 categories of SUD diagnoses, we show the effectiveness of LLMs in extracting severity information, contributing to improved risk assessment and treatment planning for SUD patients.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2024

PhysBERT: A text embedding model for physics scientific literature

The specialized language and complex concepts in physics pose significant challenges for information extraction through Natural Language Processing (NLP). Central to effective NLP applications is the text embedding model, which converts text into dense vector representations for efficient information retrieval and semantic analysis. In this work, we introduce PhysBERT, the first physics-specific text embedding model. Pre-trained on a curated corpus of 1.2 × 106 arXiv physics papers and fine-tuned with supervised data, PhysBERT outperforms leading general-purpose models on physics-specific tasks, including the effectiveness in fine-tuning for specific physics subdomains.

Hellert, Thorsten (ORCID:0000000227970926)↗

DOE OSTI2025

Dynamic in-context learning with conversational models for data extraction and materials property prediction

The advent of natural language processing and large language models (LLMs) has revolutionized the extraction of data from unstructured scholarly papers. However, ensuring data trustworthiness remains a significant challenge. In this paper, we introduce PropertyExtractor, an open-source tool that leverages advanced conversational LLMs such as Google gemini-pro and OpenAI gpt-4, blends zero-shot with few-shot in-context learning, and employs engineered prompts for the dynamic refinement of structured information hierarchies—enabling autonomous, efficient, scalable, and accurate identification, extraction, and verification of material property data. Our tests on material data demonstrate precision and recall that exceed 95% with an error rate of ∼9%, highlighting the effectiveness and versatility of the toolkit. Finally, databases for 2D material thicknesses, a critical parameter for device integration, and energy bandgap values are developed using PropertyExtractor. In particular, for the thickness database, the rapid evolution of the field has outpaced both experimental measurements and computational methods, creating a significant data gap. Our work addresses this gap and showcases the potential of PropertyExtractor as a reliable and efficient tool for the autonomous generation of various material property databases, advancing the field.

Ekuma, Chinedu E. (ORCID:0000000258527556)↗

DOE OSTI2024

Harnessing large language models’ zero-shot and few-shot learning capabilities for regulatory research

Abstract Large language models (LLMs) are sophisticated AI-driven models trained on vast sources of natural language data. They are adept at generating responses that closely mimic human conversational patterns. One of the most notable examples is OpenAI's ChatGPT, which has been extensively used across diverse sectors. Despite their flexibility, a significant challenge arises as most users must transmit their data to the servers of companies operating these models. Utilizing ChatGPT or similar models online may inadvertently expose sensitive information to the risk of data breaches. Therefore, implementing LLMs that are open source and smaller in scale within a secure local network becomes a crucial step for organizations where ensuring data privacy and protection has the highest priority, such as regulatory agencies. As a feasibility evaluation, we implemented a series of open-source LLMs within a regulatory agency’s local network and assessed their performance on specific tasks involving extracting relevant clinical pharmacology information from regulatory drug labels. Our research shows that some models work well in the context of few- or zero-shot learning, achieving performance comparable, or even better than, neural network models that needed thousands of training samples. One of the models was selected to address a real-world issue of finding intrinsic factors that affect drugs' clinical exposure without any training or fine-tuning. In a dataset of over 700 000 sentences, the model showed a 78.5% accuracy rate. Our work pointed to the possibility of implementing open-source LLMs within a secure local network and using these models to perform various natural language processing tasks when large numbers of training examples are unavailable.

Biochemistry & Molecular Biology↗

DOE OSTI2025

Data imbalance in drug response prediction: multi-objective optimization approach in deep learning setting

Abstract Drug response prediction (DRP) methods tackle the complex task of associating the effectiveness of small molecules with the specific genetic makeup of the patient. Anti-cancer DRP is a particularly challenging task requiring costly experiments as underlying pathogenic mechanisms are broad and associated with multiple genomic pathways. The scientific community has exerted significant efforts to generate public drug screening datasets, giving a path to various machine learning models that attempt to reason over complex data space of small compounds and biological characteristics of tumors. However, the data depth is still lacking compared to application domains like computer vision or natural language processing domains, limiting current learning capabilities. To combat this issue and improves the generalizability of the DRP models, we are exploring strategies that explicitly address the imbalance in the DRP datasets. We reframe the problem as a multi-objective optimization across multiple drugs to maximize deep learning model performance. We implement this approach by constructing Multi-Objective Optimization Regularized by Loss Entropy loss function and plugging it into a Deep Learning model. We demonstrate the utility of proposed drug discovery methods and make suggestions for further potential application of the work to achieve desirable outcomes in the healthcare field.

Biochemistry & Molecular Biology↗

DOE OSTI2025

Domain-specific text embedding model for accelerator physics

Accelerator physics presents unique challenges for natural language processing (NLP) due to its specialized terminology and complex concepts. A key component in overcoming these challenges is the development of robust text embedding models that transform textual data into dense vector representations, facilitating efficient information retrieval and semantic understanding. In this work, we introduce AccPhysBERT, a sentence embedding model fine-tuned specifically for accelerator physics. Our model demonstrates superior performance across a range of downstream NLP tasks, surpassing existing models in capturing the domain-specific nuances of the field. We further showcase its practical applications, including semantic paper-reviewer matching and integration into retrieval-augmented generation systems, highlighting its potential to enhance information retrieval and knowledge discovery in accelerator physics. Published by the American Physical Society 2025

Hellert, Thorsten (ORCID:0000000227970926)↗

DOE OSTI2024

Forecast of Wildfire Potential Across California USA Using a Transformer

Wildfires are a major issue facing the United States, a matter further exacerbated by an ever-changing climate. In California alone, wildfires are responsible for billions of dollars in damages and take lives each year. Accurately predicting fire danger conditions allows preparation awareness before wildfires start. Transformers are a class of deep learning models designed to identify patterns in sequential datasets. In recent years, transformers have gained popularity through their impressive performance in natural language processing and other applications of signal recognition. This analysis demonstrates the ability of a transformer with a residual connection to forecast fire danger potential over the state of California. Wildland fire potential index (WFPI) maps collected from the US Geological Survey database from January 1st 2020 to December 31st 2023 were used to tune, train and evaluate the transformer. Meteorological inputs (provided by Daymet daily weather and climatological summaries), the normalized difference vegetation index (NDVI) (calculated from the Moderate Resolution Imaging Spectroradiometer (MODIS)), and outputs from the Scott and Burgman fire behavior fuel models (to characterize maps of fuel types), were used as inputs. Our results show that a transformer can effectively emulate the US Forest Service modeled WFPI maps of California USA for four week long forecasts over the month of July, 2023, with correlations ranging from 0.85 – 0.98.

Limber, Russell [ORNL]↗

DOE OSTI2025

Exploring Capability of Multimodal Foundation Model for Image-based Fault Detection of Photovoltaic Modules

Multimodal Foundation Model (MFM), like ChatGPT and Gemini, have emerged as powerful tools for their exceptional natural language processing capabilities and their emerging potential in image analysis. This paper investigates the application of MFMs for photovoltaic (PV) fault detection through image analysis, focusing on ChatGPT 4.0 and Gemini 1.5 Pro. Three types of PV images and the corresponding common PV faults are detected: bird droppings using visible images, cell cracks via electroluminescence (EL) images, and hotspots using infrared (IR) images. Among the two models, Gemini 1.5 Pro demonstrated superior performance, achieving near-perfect results with an average F1 score of 0.97, consistently outperforming ChatGPT 4.0 in accuracy and reliability. Unlike traditional machine learning (ML) models, MFMs can operate in a zero shot manner that does not require additional training by the user, and the input images are not limited by size, angle, scope, or PV technology. The strong adaptability and user-friendliness make MFM a promising tool for analyzing PV images and advancing health monitoring for PV modules.

Li, Baojie↗

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↗