Bo Xu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Bo Xu has a total of 60 co-authors. Here are the top co-authors by number of shared publications:
- Kai Peng
Huazhong University of Science and Technology
6 shared publications - Menglan Hu
Huazhong University of Science and Technology
3 shared publications - Zhiyong Zha
State Grid Hubei Information & Telecommunication Company, Wuhan 430048, China
3 shared publications - Benkuan Zhou
Hubei Key Laboratory of Internet of Intelligence, School of Electronic Information and Communications, Huazhong University of Science and Technology, Wuhan 430074, China
2 shared publications - Bin Luo
Hubei Huazhong Electric Power Technology Development Co., Ltd., Wuhan 430079, China
2 shared publications - Junfei Wang
Hubei Key Laboratory of Internet of Intelligence, School of Electronic Information and Communications, Huazhong University of Science and Technology, Wuhan 430074, China
2 shared publications
Latest Publications
- Open Access
Cooperative UAV Swarm Data Collection: Global Coverage Memory and Neuro-Physical Action Control with FSCM-MAPPO
The increasing deployment of Unmanned Aerial Vehicle (UAV) swarms in Internet of Things (IoT) systems enables large-scale cooperative data collection. However, cooperative data collection by UAV swarms remains difficult due to sparse rewards, redundant exploration, inefficient coordination, and the need to refine discrete policy proposals before environment execution. To address these limitations, this paper studies cooperative UAV target collection in randomized simulated environments with observable target and obstacle locations and proposes a Full Spatial Coverage Memory Multi-Agent Proximal Policy Optimization (FSCM-MAPPO), a cooperative data collection algorithm integrating global perception grid with dynamic shared memory, a dense reward shaping mechanism, and a neuro-physical hybrid action controller. Specifically, a global perception grid with dynamic shared memory provides explicit spatial memory and coverage awareness, in which the global coverage information is complementary to each UAV’s local observation, thereby alleviating reward sparsity and redundant search. A neuro-physical hybrid action controller refines policy outputs through local geometric screening, velocity-based intent refinement, and discrete action selection. In an enhanced Multi-Agent Particle Environment with dynamic energy constraints, FSCM-MAPPO attains full-task success and target collection across five independent seeds under deterministic action selection, outperforming the tested MARL baselines. Ablations further show that coverage memory improves visitation and reduces repeated visits, while dense shaping and hybrid control aid task completion. Furthermore, fixed-policy comparisons reveal that hybrid control reduces accumulated contact exposure but increases contact-entry events, indicating a trade-off between task completion and contact behavior.Open AccessCooperative UAV Swarm Data Collection: Global Coverage Memory and Neuro-Physical Action Control with FSCM-MAPPO
The increasing deployment of Unmanned Aerial Vehicle (UAV) swarms in Internet of Things (IoT) systems enables large-scale cooperative data collection. However, cooperative data collection by UAV swarms remains difficult due to sparse rewards, redundant exploration, inefficient coordination, and the need to refine discrete policy proposals before environment execution. To address these limitations, this paper studies cooperative UAV target collection in randomized simulated environments with observable target and obstacle locations and proposes a Full Spatial Coverage Memory Multi-Agent Proximal Policy Optimization (FSCM-MAPPO), a cooperative data collection algorithm integrating global perception grid with dynamic shared memory, a dense reward shaping mechanism, and a neuro-physical hybrid action controller. Specifically, a global perception grid with dynamic shared memory provides explicit spatial memory and coverage awareness, in which the global coverage information is complementary to each UAV’s local observation, thereby alleviating reward sparsity and redundant search. A neuro-physical hybrid action controller refines policy outputs through local geometric screening, velocity-based intent refinement, and discrete action selection. In an enhanced Multi-Agent Particle Environment with dynamic energy constraints, FSCM-MAPPO attains full-task success and target collection across five independent seeds under deterministic action selection, outperforming the tested MARL baselines. Ablations further show that coverage memory improves visitation and reduces repeated visits, while dense shaping and hybrid control aid task completion. Furthermore, fixed-policy comparisons reveal that hybrid control reduces accumulated contact exposure but increases contact-entry events, indicating a trade-off between task completion and contact behavior. - Open Access
Permutation-Equivariant Graph Reinforcement Learning for Thermal-Aware Microservice Scheduling in Co-Packaged Optics Data Centers
As data-center interconnects move to co-packaged optics (CPO), high-power application-specific integrated circuits (ASICs) and heat-sensitive optical engines share a single interposer, and the resulting intra-module thermal coupling overwhelms conventional schedulers. Thermal-aware microservice directed acyclic graph (DAG) scheduling on CPO modules is recast here as a question of symmetry. Whereas a homogeneous-graph policy assumes the full node-permutation symmetry, that symmetry is broken twice: by the distinct task and processor node types, and by the asymmetric ASIC–engine coupling. We therefore propose a heterogeneous-graph Proximal Policy Optimization (PPO) scheduler in which the placement head is permutation-equivariant, the delay head is permutation-invariant, and the parameterization stays invariant to the processor count N. Because these symmetries hold by construction, the policy transfers zero-shot across module sizes. Heterogeneous edge typing and the resistor–capacitor (RC) coupling edge attribute are isolated by a six-test ablation chain. Evaluated on the Alibaba 2021 microservice trace across all module sizes and ambient regimes under the standard auto-cool budget, the proposed scheduler cuts the thermal-violation rate from roughly under the Heterogeneous Earliest-Finish-Time (HEFT) heuristic to about ; at the hot operating point it lowers peak temperature by about and raises DAG completion from about to , with the rare residual violations most frequent in the extreme-ambient band. With the env auto-cool budget disabled, a controlled single-axis comparison shows that removing the RC-coupling edge attribute raises the violation rate by over an order of magnitude, isolating its contribution. A single parameter set serves every N without retraining.Open AccessPermutation-Equivariant Graph Reinforcement Learning for Thermal-Aware Microservice Scheduling in Co-Packaged Optics Data Centers
As data-center interconnects move to co-packaged optics (CPO), high-power application-specific integrated circuits (ASICs) and heat-sensitive optical engines share a single interposer, and the resulting intra-module thermal coupling overwhelms conventional schedulers. Thermal-aware microservice directed acyclic graph (DAG) scheduling on CPO modules is recast here as a question of symmetry. Whereas a homogeneous-graph policy assumes the full node-permutation symmetry, that symmetry is broken twice: by the distinct task and processor node types, and by the asymmetric ASIC–engine coupling. We therefore propose a heterogeneous-graph Proximal Policy Optimization (PPO) scheduler in which the placement head is permutation-equivariant, the delay head is permutation-invariant, and the parameterization stays invariant to the processor count N. Because these symmetries hold by construction, the policy transfers zero-shot across module sizes. Heterogeneous edge typing and the resistor–capacitor (RC) coupling edge attribute are isolated by a six-test ablation chain. Evaluated on the Alibaba 2021 microservice trace across all module sizes and ambient regimes under the standard auto-cool budget, the proposed scheduler cuts the thermal-violation rate from roughly under the Heterogeneous Earliest-Finish-Time (HEFT) heuristic to about ; at the hot operating point it lowers peak temperature by about and raises DAG completion from about to , with the rare residual violations most frequent in the extreme-ambient band. With the env auto-cool budget disabled, a controlled single-axis comparison shows that removing the RC-coupling edge attribute raises the violation rate by over an order of magnitude, isolating its contribution. A single parameter set serves every N without retraining. - Open Access
PromptTrace: A Fine-Grained Prompt Stealing Attack via CLIP-Guided Beam Search for Text-to-Image Models
The inherent semantic symmetry and cross-modal alignment between textual prompts and generated images have fueled the success of text-to-image (T2I) generation. However, this strong correlation also introduces security vulnerabilities, specifically prompt stealing attacks, where valuable prompts are reverse-engineered from images. In this paper, we address the challenge of information asymmetry in black-box attack scenarios and propose PromptTrace, a fine-grained prompt stealing framework via Contrastive Language-Image Pre-training (CLIP)-guidedbeam search. Unlike existing methods that rely on single-stage generation, PromptTrace structurally decomposes prompt reconstruction into subject generation, modifier extraction, and iterative search optimization to effectively restore the visual–textual correspondence. By leveraging a CLIP-guided beam search strategy, our method progressively optimizes candidate prompts based on image–text similarity feedback, ensuring the stolen prompt achieves high fidelity in both semantic intent and stylistic representation. Extensive evaluations across multiple datasets and T2I models demonstrate that PromptTrace outperforms existing methods, highlighting the feasibility of exploiting cross-modal symmetry for attacks and underscoring the urgent need for defense mechanisms in the T2I ecosystem.
Open AccessPromptTrace: A Fine-Grained Prompt Stealing Attack via CLIP-Guided Beam Search for Text-to-Image Models
The inherent semantic symmetry and cross-modal alignment between textual prompts and generated images have fueled the success of text-to-image (T2I) generation. However, this strong correlation also introduces security vulnerabilities, specifically prompt stealing attacks, where valuable prompts are reverse-engineered from images. In this paper, we address the challenge of information asymmetry in black-box attack scenarios and propose PromptTrace, a fine-grained prompt stealing framework via Contrastive Language-Image Pre-training (CLIP)-guidedbeam search. Unlike existing methods that rely on single-stage generation, PromptTrace structurally decomposes prompt reconstruction into subject generation, modifier extraction, and iterative search optimization to effectively restore the visual–textual correspondence. By leveraging a CLIP-guided beam search strategy, our method progressively optimizes candidate prompts based on image–text similarity feedback, ensuring the stolen prompt achieves high fidelity in both semantic intent and stylistic representation. Extensive evaluations across multiple datasets and T2I models demonstrate that PromptTrace outperforms existing methods, highlighting the feasibility of exploiting cross-modal symmetry for attacks and underscoring the urgent need for defense mechanisms in the T2I ecosystem.
- Open Access
FedRP: Region-Specific Personalized Identification for Large-Scale IoT Systems
The widespread adoption of Internet of Things (IoT) technology has significantly expanded the scale at which devices are connected, posing new challenges to maintaining symmetry in network management. Traditional centralized identification architectures adopt a symmetric processing paradigm in which all device data are uniformly transmitted to the cloud for processing. However, this rigid symmetric structure fails to accommodate the asymmetric distribution typical of IoT edge devices. To address these challenges, this paper proposes an asymmetric identification framework based on cloud–edge collaboration, exploring a high-performance, resource-efficient, and privacy-preserving solution for IoT device identification. The proposed region-specific personalized algorithm (FedRP) introduces a region-specific, personalized identification approach grounded in federated learning principles. Firstly, FedRP leverages a decentralized processing framework to enhance data security by processing data locally. Secondly, it employs a personalized federated learning framework to optimize local models, thus improving identification accuracy and effectiveness. Finally, FedRP strategically separates the personalized parameters of transformer-based blocks from shared parameters and selectively transmits them, reducing the burden on network resources. Comprehensive comparative experiments demonstrate the efficacy of the proposed approach for large-scale IoT environments, which are characterized by numerous devices and complex network conditions.
Open AccessFedRP: Region-Specific Personalized Identification for Large-Scale IoT Systems
The widespread adoption of Internet of Things (IoT) technology has significantly expanded the scale at which devices are connected, posing new challenges to maintaining symmetry in network management. Traditional centralized identification architectures adopt a symmetric processing paradigm in which all device data are uniformly transmitted to the cloud for processing. However, this rigid symmetric structure fails to accommodate the asymmetric distribution typical of IoT edge devices. To address these challenges, this paper proposes an asymmetric identification framework based on cloud–edge collaboration, exploring a high-performance, resource-efficient, and privacy-preserving solution for IoT device identification. The proposed region-specific personalized algorithm (FedRP) introduces a region-specific, personalized identification approach grounded in federated learning principles. Firstly, FedRP leverages a decentralized processing framework to enhance data security by processing data locally. Secondly, it employs a personalized federated learning framework to optimize local models, thus improving identification accuracy and effectiveness. Finally, FedRP strategically separates the personalized parameters of transformer-based blocks from shared parameters and selectively transmits them, reducing the burden on network resources. Comprehensive comparative experiments demonstrate the efficacy of the proposed approach for large-scale IoT environments, which are characterized by numerous devices and complex network conditions.
- Open Access
Towards Faithful Local Explanations: Leveraging SVM to Interpret Black-Box Machine Learning Models
Although machine learning (ML) models are widely used in many fields, their prediction processes are often hard to understand. This lack of transparency makes it harder for people to trust them, especially in high-stakes fields like healthcare and finance. Human-interpretable explanations for model predictions are crucial in these contexts. While existing local interpretation methods have been proposed, many suffer from low local fidelity, instability, and limited effectiveness when applied to highly nonlinear models. This paper presents SVM-X, a model-agnostic local explanation approach designed to address these challenges. By leveraging the inherent symmetry of the SVM hyperplane, SVM-X precisely captures the local decision boundaries of complex nonlinear models, providing more accurate and stable explanations. Experimental evaluations on the UCI Adult dataset, the Bank Marketing dataset, and the Amazon Product Review dataset demonstrate that SVM-X consistently outperforms state-of-the-art methods like LIME and LEMNA. Notably, SVM-X achieves up to a 27.2% improvement in accuracy. Our work introduces a reliable and interpretable framework for understanding machine learning predictions, offering a promising new direction for future research.
Open AccessTowards Faithful Local Explanations: Leveraging SVM to Interpret Black-Box Machine Learning Models
Although machine learning (ML) models are widely used in many fields, their prediction processes are often hard to understand. This lack of transparency makes it harder for people to trust them, especially in high-stakes fields like healthcare and finance. Human-interpretable explanations for model predictions are crucial in these contexts. While existing local interpretation methods have been proposed, many suffer from low local fidelity, instability, and limited effectiveness when applied to highly nonlinear models. This paper presents SVM-X, a model-agnostic local explanation approach designed to address these challenges. By leveraging the inherent symmetry of the SVM hyperplane, SVM-X precisely captures the local decision boundaries of complex nonlinear models, providing more accurate and stable explanations. Experimental evaluations on the UCI Adult dataset, the Bank Marketing dataset, and the Amazon Product Review dataset demonstrate that SVM-X consistently outperforms state-of-the-art methods like LIME and LEMNA. Notably, SVM-X achieves up to a 27.2% improvement in accuracy. Our work introduces a reliable and interpretable framework for understanding machine learning predictions, offering a promising new direction for future research.
- Open Access
- 10 April 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (8) , 4177
Neighborhood Deviation Attack Against In-Context Learning
In-context learning (ICL) enables large language models (LLMs) to adapt to new tasks using only a few examples, without requiring fine-tuning. However, the new privacy and security risks brought about by this increasing capability have not received enough attention, and there is a lack of research on this issue. In this work, we propose a novel membership inference attack (MIA) method, termed Neighborhood Deviation Attack, specifically designed to evaluate the privacy risks of LLMs in ICL. Unlike traditional MIA methods, our approach does not require access to model parameters and instead relies solely on analyzing the model’s output behavior. We first generate neighborhood prefixes for target samples and use the LLM, conditioned on ICL examples, to complete the text. We then compute the deviation between the original and completed texts and infer membership based on these deviations. We conduct experiments on three datasets and three LLMs and further explore the influence of key hyperparameters on the method’s performance and their underlying reasons. Experimental results show that our method is significantly better than the comparative methods in terms of stability and achieves better accuracy in most cases. Furthermore, we discuss four potential defense strategies, including increasing the diversity of ICL examples and introducing controlled randomness in the inference process to reduce the risk of privacy leakage.
- 10 April 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (8) , 4177
Open AccessNeighborhood Deviation Attack Against In-Context Learning
In-context learning (ICL) enables large language models (LLMs) to adapt to new tasks using only a few examples, without requiring fine-tuning. However, the new privacy and security risks brought about by this increasing capability have not received enough attention, and there is a lack of research on this issue. In this work, we propose a novel membership inference attack (MIA) method, termed Neighborhood Deviation Attack, specifically designed to evaluate the privacy risks of LLMs in ICL. Unlike traditional MIA methods, our approach does not require access to model parameters and instead relies solely on analyzing the model’s output behavior. We first generate neighborhood prefixes for target samples and use the LLM, conditioned on ICL examples, to complete the text. We then compute the deviation between the original and completed texts and infer membership based on these deviations. We conduct experiments on three datasets and three LLMs and further explore the influence of key hyperparameters on the method’s performance and their underlying reasons. Experimental results show that our method is significantly better than the comparative methods in terms of stability and achieves better accuracy in most cases. Furthermore, we discuss four potential defense strategies, including increasing the diversity of ICL examples and introducing controlled randomness in the inference process to reduce the risk of privacy leakage.
- Open Access
Optimizing Microservice Deployment in Edge Computing with Large Language Models: Integrating Retrieval Augmented Generation and Chain of Thought Techniques
Large Language Models (LLMs) have demonstrated impressive capabilities in autogenerating code based on natural language instructions provided by humans. We observed that in the microservice models of edge computing, the problem of deployment latency optimization can be transformed into an NP-hard mathematical optimization problem. However, in the real world, deployment strategies at the edge often require immediate updates, while human-engineered code tends to be lagging. To bridge this gap, we innovatively integrated LLMs into the decision-making process for microservice deployment. Initially, we constructed a private Retrieval Augmented Generation (RAG) database containing prior knowledge. Subsequently, we employed meticulously designed step-by-step inductive instructions and used the chain of thought (CoT) technique to enable the LLM to learn, reason, reflect, and regenerate. We decomposed the microservice deployment latency optimization problem into a collection of granular sub-problems (described in natural language), progressively providing instructions to the fine-tuned LLM to generate corresponding code blocks. The generated code blocks underwent integration and consistency assessment. Additionally, we prompted the LLM to generate code without the use of the RAG database for comparative analysis. We executed the aforementioned code and comparison algorithm under identical operational environments and simulation parameters, conducting rigorous result analysis. Our fine-tuned model significantly reduced latencies by 22.8% in handling surges in request flows, 37.8% in managing complex microservice types, and 39.5% in processing increased network nodes compared to traditional algorithms. Moreover, our approach demonstrated marked improvements in latency performance over LLMs not utilizing RAG technology and reinforcement learning algorithms reported in other literature. The use of LLMs also highlights the concept of symmetry, as the symmetrical structure of input-output relationships in microservice deployment models aligns with the LLM’s inherent ability to process and generate balanced and optimized code. Symmetry in this context allows for more efficient resource allocation and reduces redundant operations, further enhancing the model’s effectiveness. We believe that LLMs hold substantial potential in optimizing microservice deployment models.
Open AccessOptimizing Microservice Deployment in Edge Computing with Large Language Models: Integrating Retrieval Augmented Generation and Chain of Thought Techniques
Large Language Models (LLMs) have demonstrated impressive capabilities in autogenerating code based on natural language instructions provided by humans. We observed that in the microservice models of edge computing, the problem of deployment latency optimization can be transformed into an NP-hard mathematical optimization problem. However, in the real world, deployment strategies at the edge often require immediate updates, while human-engineered code tends to be lagging. To bridge this gap, we innovatively integrated LLMs into the decision-making process for microservice deployment. Initially, we constructed a private Retrieval Augmented Generation (RAG) database containing prior knowledge. Subsequently, we employed meticulously designed step-by-step inductive instructions and used the chain of thought (CoT) technique to enable the LLM to learn, reason, reflect, and regenerate. We decomposed the microservice deployment latency optimization problem into a collection of granular sub-problems (described in natural language), progressively providing instructions to the fine-tuned LLM to generate corresponding code blocks. The generated code blocks underwent integration and consistency assessment. Additionally, we prompted the LLM to generate code without the use of the RAG database for comparative analysis. We executed the aforementioned code and comparison algorithm under identical operational environments and simulation parameters, conducting rigorous result analysis. Our fine-tuned model significantly reduced latencies by 22.8% in handling surges in request flows, 37.8% in managing complex microservice types, and 39.5% in processing increased network nodes compared to traditional algorithms. Moreover, our approach demonstrated marked improvements in latency performance over LLMs not utilizing RAG technology and reinforcement learning algorithms reported in other literature. The use of LLMs also highlights the concept of symmetry, as the symmetrical structure of input-output relationships in microservice deployment models aligns with the LLM’s inherent ability to process and generate balanced and optimized code. Symmetry in this context allows for more efficient resource allocation and reduces redundant operations, further enhancing the model’s effectiveness. We believe that LLMs hold substantial potential in optimizing microservice deployment models.
- Open Access
- 25 September 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14 (19) , 8656
Energy-Efficient Joint Partitioning and Offloading for Delay-Sensitive CNN Inference in Edge Computing
With the development of deep learning foundation model technology, the types of computing tasks have become more complex, and the computing resources and memory required for these tasks have also become more substantial. Since it has long been revealed that task offloading in cloud servers has many drawbacks, such as high communication delay and low security, task offloading is mostly carried out in the edge servers of the Internet of Things (IoT) network. However, edge servers in IoT networks are characterized by tight resource constraints and often the dynamic nature of data sources. Therefore, the question of how to perform task offloading of deep learning foundation model services on edge servers has become a new research topic. However, the existing task offloading methods either can not meet the requirements of massive CNN architecture or require a lot of communication overhead, leading to significant delays and energy consumption. In this paper, we propose a parallel partitioning method based on matrix convolution to partition foundation model inference tasks, which partitions large CNN inference tasks into subtasks that can be executed in parallel to meet the constraints of edge devices with limited hardware resources. Then, we model and mathematically express the problem of task offloading. In a multi-edge-server, multi-user, and multi-task edge-end system, we propose a task-offloading method that balances the tradeoff between delay and energy consumption. It adopts a greedy algorithm to optimize task-offloading decisions and terminal device transmission power to maximize the benefits of task offloading. Finally, extensive experiments verify the significant and extensive effectiveness of our algorithm.
- 25 September 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14 (19) , 8656
Open AccessEnergy-Efficient Joint Partitioning and Offloading for Delay-Sensitive CNN Inference in Edge Computing
With the development of deep learning foundation model technology, the types of computing tasks have become more complex, and the computing resources and memory required for these tasks have also become more substantial. Since it has long been revealed that task offloading in cloud servers has many drawbacks, such as high communication delay and low security, task offloading is mostly carried out in the edge servers of the Internet of Things (IoT) network. However, edge servers in IoT networks are characterized by tight resource constraints and often the dynamic nature of data sources. Therefore, the question of how to perform task offloading of deep learning foundation model services on edge servers has become a new research topic. However, the existing task offloading methods either can not meet the requirements of massive CNN architecture or require a lot of communication overhead, leading to significant delays and energy consumption. In this paper, we propose a parallel partitioning method based on matrix convolution to partition foundation model inference tasks, which partitions large CNN inference tasks into subtasks that can be executed in parallel to meet the constraints of edge devices with limited hardware resources. Then, we model and mathematically express the problem of task offloading. In a multi-edge-server, multi-user, and multi-task edge-end system, we propose a task-offloading method that balances the tradeoff between delay and energy consumption. It adopts a greedy algorithm to optimize task-offloading decisions and terminal device transmission power to maximize the benefits of task offloading. Finally, extensive experiments verify the significant and extensive effectiveness of our algorithm.
- Open Access
- 14 August 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14 (16) , 7118
Targeted Training Data Extraction—Neighborhood Comparison-Based Membership Inference Attacks in Large Language Models
A large language model refers to a deep learning model characterized by extensive parameters and pretraining on a large-scale corpus, utilized for processing natural language text and generating high-quality text output. The increasing deployment of large language models has brought significant attention to their associated privacy and security issues. Recent experiments have demonstrated that training data can be extracted from these models due to their memory effect. Initially, research on large language model training data extraction focused primarily on non-targeted methods. However, following the introduction of targeted training data extraction by Carlini et al., prefix-based extraction methods to generate suffixes have garnered considerable interest, although current extraction precision remains low. This paper focuses on the targeted extraction of training data, employing various methods to enhance the precision and speed of the extraction process. Building on the work of Yu et al., we conduct a comprehensive analysis of the impact of different suffix generation methods on the precision of suffix generation. Additionally, we examine the quality and diversity of text generated by various suffix generation strategies. The study also applies membership inference attacks based on neighborhood comparison to the extraction of training data in large language models, conducting thorough evaluations and comparisons. The effectiveness of membership inference attacks in extracting training data from large language models is assessed, and the performance of different membership inference attacks is compared. Hyperparameter tuning is performed on multiple parameters to enhance the extraction of training data. Experimental results indicate that the proposed method significantly improves extraction precision compared to previous approaches.
- 14 August 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14 (16) , 7118
Open AccessTargeted Training Data Extraction—Neighborhood Comparison-Based Membership Inference Attacks in Large Language Models
A large language model refers to a deep learning model characterized by extensive parameters and pretraining on a large-scale corpus, utilized for processing natural language text and generating high-quality text output. The increasing deployment of large language models has brought significant attention to their associated privacy and security issues. Recent experiments have demonstrated that training data can be extracted from these models due to their memory effect. Initially, research on large language model training data extraction focused primarily on non-targeted methods. However, following the introduction of targeted training data extraction by Carlini et al., prefix-based extraction methods to generate suffixes have garnered considerable interest, although current extraction precision remains low. This paper focuses on the targeted extraction of training data, employing various methods to enhance the precision and speed of the extraction process. Building on the work of Yu et al., we conduct a comprehensive analysis of the impact of different suffix generation methods on the precision of suffix generation. Additionally, we examine the quality and diversity of text generated by various suffix generation strategies. The study also applies membership inference attacks based on neighborhood comparison to the extraction of training data in large language models, conducting thorough evaluations and comparisons. The effectiveness of membership inference attacks in extracting training data from large language models is assessed, and the performance of different membership inference attacks is compared. Hyperparameter tuning is performed on multiple parameters to enhance the extraction of training data. Experimental results indicate that the proposed method significantly improves extraction precision compared to previous approaches.
- 26 March 2024
- Published by Springer Nature in Journal of Grid Computing
- Vol. 22 (2) , 1-19
On the Joint Design of Microservice Deployment and Routing in Cloud Data Centers
- 26 March 2024
- Published by Springer Nature in Journal of Grid Computing
- Vol. 22 (2) , 1-19
On the Joint Design of Microservice Deployment and Routing in Cloud Data Centers