Generation Strategy Research Articles

A generative adversarial learning strategy for spatial inspection of compaction quality

Reliable prediction methods play a crucial role in enhancing the compaction quality and implementing the intelligent compaction (IC). The predictive performance of contemporary data-driven models is significantly influenced by the quantity and quality of training data. The scarcity and high dimensionality of field compaction data may lead to overfitting issues in prediction models. In this study, an innovative methodology, spatial trajectory condition generative adversarial networks (STC-GAN), is proposed to synthesize IC datasets by considering the construction procedures. The dataset expansion is driven both by the non-attentional algorithm and the engineering practice. The augmented dataset is used to train the extreme gradient boosting (XGBoost) for estimating the compaction quality. The model accuracy is consistently enhanced by adopting the STC-GAN augmented datasets. The optimal model is achieved after a 2.5-fold expansion by performing the sensitivity study. The proposed STC-GAN-PSO-XGBoost is a physical-based intelligent model, enhancing the reliability of data augmentation and the compaction evaluation model. This method is beneficial to minimize the amount of field tests, enabling the continuous evaluation of compaction quality for IC.

Strategi Komunikasi Pendidikan untuk Meningkatkan Kualitas Pendidikan di Era Gen Z

This study discusses educational communication strategies to improve the quality of education in the Generation Z era, which requires a communication approach that suits the unique characteristics influenced by technology. Using a qualitative approach and literature analysis, the research aims to identify effective educational communication strategies for Generation Z. Key findings include visual preferences, personalization, interactivity, problem-based learning, immersive technology, digital mentoring, flexibility, and social responsibility. Recommended strategies include visual content, adaptive learning, online discussions, VR/AR, digital mentoring, mobile learning, social projects, gamification, and multi-platform communication. It is important to understand the characteristics of Generation Z and adapt in relevant and effective ways, as well as create education that is technologically advanced and of great value. Implementation suggestions include digital teacher training, online cultural diplomacy, and educational media policies. This research provides valuable guidance for the development of educational strategies that suit the needs of Z Generation.

DynER: Optimized Test Case Generation for Representational State Transfer (REST)ful Application Programming Interface (API) Fuzzers Guided by Dynamic Error Responses

Modern web services widely provide RESTful APIs for clients to access their functionality programmatically. Fuzzing is an emerging technique for ensuring the reliability of RESTful APIs. However, the existing RESTful API fuzzers repeatedly generate invalid requests due to unawareness of errors in the invalid tested requests and lack of effective strategy to generate legal value for the incorrect parameters. Such limitations severely hinder the fuzzing performance. In this paper, we propose DynER, a new test case generation method guided by dynamic error responses during fuzzing. DynER designs two strategies of parameter value generation for purposefully revising the incorrect parameters of invalid tested requests to generate new test requests. The strategies are, respectively, based on prompting Large Language Model (LLM) to understand the semantics information in error responses and actively accessing API-related resources. We apply DynER to the state-of-the-art fuzzer RESTler and implement DynER-RESTler. DynER-RESTler outperforms foREST on two real-world RESTful services, WordPress and GitLab with a 41.21% and 26.33% higher average pass rate for test requests and a 12.50% and 22.80% higher average number of unique request types successfully tested, respectively. The experimental results demonstrate that DynER significantly improves the effectiveness of test cases and fuzzing performance. Additionally, DynER-RESTler finds three new bugs.

PIS-Net: Efficient weakly supervised instance segmentation network based on annotated points for rice field weed identification

Weed damage in rice fields is one of the main causes of reduced rice yields and quality. Accurate and efficient weed identification is the prerequisite for realizing intelligent and precise weeding in paddies. Recently, Vision Transformers (ViTs) have emerged with superior performance on computer vision tasks compared to the convolutional neural network (CNN)-based models. However, the lack of fully labeled weed datasets hinders the potential application of deep learning models in weed identification. To address the above issues, this study customizes a novel point-supervised instance segmentation network (PIS-Net) for weakly supervised instance segmentation of weeds in rice fields. More correctly, we first propose a novel instance segmentation point labeling scheme that utilizes randomly generated annotation points within each instance, aiming to decrease both labeling time and difficulty. Additionally, to make optimal use of point labels, this study puts forth a mask generation strategy based on the adaptive selection of pyramid levels. In this sense, the network model can flexibly choose the pyramid level expected to generate the most suitable instance mask based on the network's reliability. Finally, we establish the pseudo label refinement network (PLR-Net) to refine rough instance masks. The proposed PIS-Net utilizes 13 randomly generated annotation points for each instance, yet achieving an AP of 38.5 and an AP50 of 68.3, which is superior to the baseline mask-R-CNN with an AP of 8.2 and AP50 of 6.9, achieving 90 % fully supervised performance. This method effectively utilizes point labels, annotated with high efficiency, as a robust source of weak supervision to address challenges in weed data annotation and the low accuracy of existing weakly supervised models. Experiments show that the point annotation scheme of the PIS-Net is faster than full-object mask annotation, and the AP is also higher than the current semi-supervised weed segmentation model, enjoying high potentials in practical paddy fields.

Environmental impact assessment of different power generation strategies in Oman: A comparative life-cycle analysis

This paper presents a comparative environmental impact assessment considering different power generation strategies in Oman. The Life-Cycle Assessment (LCA) methodology, and the OpenLCA tool, were employed in carrying out comprehensive analysis, and evaluation of the various environmental aspects in filling the research gap, by replacing conventional diesel generators with natural gas alternatives in Oman. The obtained results indicate that utilizing natural gas significantly reduces environmental impacts, including a decrease in global warming potential to 2.27 million kg CO2 eq, fossil fuel depletion to 34.5 million kg oil-eq, and ozone depletion to 0.13 kg CFC-11 eq. This study would help in policy decision-making to support a potential shift in power generation system, transitioning from the current use of diesel generator sets to flared produced Natural Gas across crude oil processing plants. Thus, mitigating operational cost, and improving efficiencies, for sustainable developments. Another salient part of this study is the implementation of more sustainable energy practices that supports the broader application of LCA in evaluating industrial environmental impacts. Furthermore, the results obtained in this work are in line with recent global climate change commitments and Oman's Vision 2040, in achieving cleaner energy sources to minimize environmental harm in the oil and gas sector.

Research on fault diagnosis of multi-mode electromechanical compound transmission system for hybrid electric vehicle based on global analytical redundancy relations

In order to effectively monitor the operational status and detect faults within a hybrid electric vehicle during mode transitions, this study focuses on a specific power-split hybrid electric vehicle model known as the multimode electromechanical composite transmission system (MM-EMCTS HEV). Taking into account the actual occurrence frequency and simulatability of faults in MM-EMCTS HEV, key faults are extracted. Based on the distinctive characteristics and properties of critical system faults, a fault diagnosis framework is constructed by leveraging global analytical redundancy relations (GARRs). This framework encompasses strategies for residual generation across sensors, actuators, and controlled components, alongside residual evaluation techniques employing mode-dependent adaptive thresholds. The detectability and isolatability of the MM-EMCTS are thoroughly analyzed. Subsequently, simulation and verification of the residual responses under typical faults in the MM-EMCTS are implemented in Matlab/Simulink and Simscape. The results demonstrate the efficacy of the fault diagnosis method, rooted in global analytical redundancy relations and mode-dependent adaptive thresholds, in successfully detecting and isolating faults within multimode electromechanical composite transmission systems. Furthermore, compared to conventional fixed threshold residual evaluation approaches, this method significantly mitigates the occurrence of missed detections, thus verifying the effectiveness and superiority of the proposed method.

Intelligent power management system for optimizing load strategies in renewable generation

Intelligent power management system for optimizing load strategies in renewable generation

Elite-guided Resampling and Multi-mutation based Differential Evolution with exponential crossover for numerical optimization

In the existing Differential Evolution (DE) research, there are two main crossover schemes: exponential crossover and binomial crossover. Most researchers believe that binomial crossover is skilled in handling numerical optimization problems while exponential crossover has the advantage in handling optimization problems with certain correlation between adjacent variables. However, we discover that DE variants with exponential crossover can achieve performance comparable to advanced ones utilizing binomial crossover in optimization applications, provided that an appropriate crossover rate CR and associated parameter control are established. In this paper, an Elite-guided Resampling and Multi-mutation based DE with exponential crossover (ERM-DE) is raised to fill the blank in this field, and its main highlights are as follows: First, a novel structure with two stages is raised in our ERM-DE algorithm. The first stage is elite-guided global opposition learning based resampling, and the other is novel exponential crossover DE evolution stage. Second, a Novel Parameter Control (NPC) technique involving scale factor F and crossover rate CR is raised. Different from applying the fitness-based parameter control, the adaptation schemes of the parameters F and CR in our ERM-DE algorithm are fitness-independent. In addition, this NPC technique will also be selectively applied to the trial vector generation strategy DE/target-to-pbest/1/exp with or without optional archive. Third, a novel population Diversity Enhancement Mechanism (DEM) is proposed, which can enhance population diversity by recalculating individuals in a stagnant state. Finally, a time stamp mechanism is posed to deal with outdated inferior and historical solutions in the external archive. A Large test suite with 88 benchmark functions from CEC2013, CEC2014, and CEC2017 test suites is used for algorithm validation and the experimental results indicate that our ERM-DE demonstrates competitiveness with several state-of-the-art DE variants.

One‐Pot Biocatalytic Conversion of Chemically Inert Hydrocarbons into Chiral Amino Acids through Internal Cofactor and H2O2 Recycling

Chemically inert hydrocarbons are the primary feedstocks used in the petrochemical industry and can be converted into more intricate and valuable chemicals. However, two major challenges impede this conversion process: selective activation of C‐H bonds in hydrocarbons and systematic functionalization required to synthesize complex structures. To address these issues, we developed a multi‐enzyme cascade conversion system based on internal cofactor and H2O2 recycling to achieve the one‐pot deep conversion from heptane to chiral (S)‐2‐aminoheptanoic acid under mild conditions. First, a hydrogen‐borrowing‐cycle‐based NADH regeneration method and H2O2in situ generation and consumption strategy were applied to realize selective C‐H bond oxyfunctionalization, converting heptane into 2‐hydroxyheptanoic acid. Integrating subsequent reductive amination driven by the second hydrogen‐borrowing cycle, (S)‐2‐aminoheptanoic acid was finally accumulated at 4.57 mM with eep > 99%. Hexane, octane, 2‐methylheptane, and butylbenzene were also successfully converted into the corresponding chiral amino acids with eep > 99%. Overall, the conversion system employed internal cofactor and H2O2 recycling, with O2 as the oxidant and ammonium as the amination reagent to fulfill the enzymatic conversion from chemically inert hydrocarbons into chiral amino acids under environmentally friendly conditions, which is a highly challenging transformation in traditional organic synthesis.

One-Pot Biocatalytic Conversion of Chemically Inert Hydrocarbons into Chiral Amino Acids through Internal Cofactor and H2O2 Recycling.

Chemically inert hydrocarbons are the primary feedstocks used in the petrochemical industry and can be converted into more intricate and valuable chemicals. However, two major challenges impede this conversion process: selective activation of C-H bonds in hydrocarbons and systematic functionalization required to synthesize complex structures. To address these issues, we developed a multi-enzyme cascade conversion system based on internal cofactor and H2O2 recycling to achieve the one-pot deep conversion from heptane to chiral (S)-2-aminoheptanoic acid under mild conditions. First, a hydrogen-borrowing-cycle-based NADH regeneration method and H2O2in situ generation and consumption strategy were applied to realize selective C-H bond oxyfunctionalization, converting heptane into 2-hydroxyheptanoic acid. Integrating subsequent reductive amination driven by the second hydrogen-borrowing cycle, (S)-2-aminoheptanoic acid was finally accumulated at 4.57 mM with eep > 99%. Hexane, octane, 2-methylheptane, and butylbenzene were also successfully converted into the corresponding chiral amino acids with eep > 99%. Overall, the conversion system employed internal cofactor and H2O2 recycling, with O2 as the oxidant and ammonium as the amination reagent to fulfill the enzymatic conversion from chemically inert hydrocarbons into chiral amino acids under environmentally friendly conditions, which is a highly challenging transformation in traditional organic synthesis.

An Area-Efficient and Configurable Number Theoretic Transform Accelerator for Homomorphic Encryption

Homomorphic Encryption (HE) allows for arbitrary computation of encrypted data, offering a method for preserving privacy in cloud computations. However, efficiency remains a significant obstacle, particularly with the polynomial multiplication of large parameter sets, which occupies substantial computing and memory overhead. Prior studies proposed the use of Number Theoretic Transform (NTT) to accelerate polynomial multiplication, which proved efficient, owing to its low computational complexity. However, these efforts primarily focused on NTT designs for small parameter sets, and configurability and memory efficiency were not considered carefully. This paper focuses on designing a unified NTT/Inverse NTT (INTT) architecture with high area efficiency and configurability, which is more suitable for HE schemes. We adopt the Constant-Geometry (CG) NTT algorithm and propose a conflict-free access pattern, demonstrating a 16.7% reduction in coefficients of storage capacity compared to the state-of-the-art CG NTT design. Additionally, we propose a twiddle factor generation strategy to minimize storage for Twiddle Factors (TFs). The proposed architecture offers configurability of both compile time and runtime, allowing for the deployment of varying parallelism and parameter sets during compilation while accommodating a wide range of polynomial degrees and moduli after compilation. Experimental results on the Xilinx FPGA show that our design can achieve higher area efficiency and configurability compared with previous works. Furthermore, we explore the performance difference between precomputed TFs and online-generated TFs for the NTT architecture, aiming to show the importance of online generation-based NTT architecture in HE applications.

Resilience enhancement by line hardening for communication routing considering renewable energy sources in cyber‐physical power systems

AbstractCommunication routing of cyber‐physical power systems (CPPS) with the high penetration of renewable energy sources (RES) plays an important role in the resilience enhancement against disasters, natural and man‐made alike. Therefore, a trilevel optimization model is proposed for the CPPS resilience enhancement with the RES against extreme events. The upper‐level model identifies optimal hardening lines of both the transmission and communication networks with consideration of the communication routing constraints. The middle‐level model identifies the attacked lines to maximize load shedding of the power system. The effects of RES uncertainty and communication routings on resilience enhancement are analysed. The lower‐level model attains an optimal allocation strategy of power generation to minimize load shedding. The model is solved by the column‐and‐constraint generation algorithm. Case studies are conducted on the IEEE 14‐bus, RTS 24‐bus, and 118‐bus system. The results show that the proposed hardening strategy can effectively ensure the adaptiveness of the transmission network and communication routing to improve the resilience of CPPS to extreme events. Moreover, it is observed that the load loss and total investment cost are heavily affected without and with RES.

Healthy older adults generate transverse-plane momenta required for 90° turns while walking during the same phases of gait as used in straight-line gait

BackgroundGeneration and regulation (control) of linear and angular momentum is a challenge during turning while walking which may be exacerbated by age-related changes. In healthy older adults, little is known about how momentum is controlled during turns, especially within each phase of gait. Each phase of gait affords unique mechanical contexts to control momenta and regulate balance. In healthy young adults, we found that the transverse-plane linear and angular momenta generation strategies observed within specific phases of gait during straight-line gait were also used during turns. Therefore, in this study, we investigated whether healthy older adults shared similar momentum control strategies specific to each gait phase during straight-line gait and turns.MethodsNine healthy older adults completed straight-line gait and 90° leftward walking turns. We compared the change in transverse-plane whole-body linear and angular momentum across gait phases (left and right single and double support). We also compared the average leftward force and transverse-plane moment across gait phases.ResultsWe found that leftward linear momentum was generated most during right single support in straight-line gait and leftward turns. However, in contrast to straight-line gait, during leftward turns, average leftward force was applied across gait phases, with left single support generating significantly less leftward average force than other gait phases. Leftward angular momentum generation and average moment were greatest during left double support in both tasks. We observed some within-participant results that diverged from the group statistical findings, illustrating that although they are common, these momenta control strategies are not necessary.ConclusionsOlder adults generated transverse-plane linear and angular momentum during consistent phases of gait during straight-line gait and 90° turns, potentially indicating a shared control strategy. Understanding momentum control within each phase of gait can help design more specific targets in gait and balance training interventions.

Scale-Bridging Mechanics Transfer Enables Ultrabright Mechanoluminescent Fiber Electronics.

Mechanoluminescent (ML) fibers and textiles enable stress visualization without auxiliary power, showing great potential in wearable electronics, machine vision, and human-computer interaction. However, traditional ML devices suffer from inefficient stress transfer in soft-rigid material systems, leading to low luminescence brightness and short cycle life. Here, we propose a tendon-inspired scale-bridging mechanics transfer mechanism for ML composites, which employs molecular-scale copolymerized cross-linking and nanoscale inorganic nanoparticles as hierarchical stress transfer sites. This strategy effectively reduces the dissipation of stress in molecular chain segments and alleviates local stress concentration, increases luminescence by 9 times, and extends cycle life to more than 10,000 times. Furthermore, a scalable (kilometer-scale) anti-Plateau-Rayleigh instability manufacturing technology is developed for thermoset ML fibers, compatible with various existing textile techniques. We also demonstrate its system-level applications in motion capture, underwater interaction, etc., providing a feasible strategy for the next generation of smart visual textiles.

Estimating the actual emission cost in an annual compliance cycle: Synergistic generation and carbon trading optimization for price-taking generation companies

The annual compliance cycle of the carbon trading system gives generation companies (GenCos) the flexibility to buy allowances at preferred times rather than the same time that emissions happen. How to estimate the actual emission cost under fluctuating carbon prices and make better generation decisions become challenges for GenCos. A synergistic generation and carbon trading optimization method is proposed in this paper with a non-speculative dynamic (NSD) carbon trading strategy. The NSD strategy utilizes stochastic dynamic programming to make optimal carbon trading decisions and to provide accurate emission cost estimations for GenCos to make generation decisions. To control risks, the NSD strategy forbids intentional speculations in the carbon market but allows both buying and selling of allowances according to the yearly total emissions estimated by the generation strategy. A risk control scheme by adjusting the allowance-emission balancing period is proposed so that GenCos can smoothly shift between the traditional daily-balanced carbon trading strategy and the NSD strategy. Properties of the NSD strategy are analyzed and proved theoretically under some sufficient conditions, which makes the strategy interpretable and trustworthy. Numerical simulations verify that the proposed method can help GenCos lower their emission costs and improve their overall profits in a year.

Assessing Greenhouse Gas Emissions and Methane Production During Post-Landfill Maintenance in Kish Island: A LandGEM Simulation Approach

Background: Municipal solid waste (MSW) landfills are significant contributors to the emission of greenhouse gases, particularly methane, during the breakdown of waste materials. Researchers have undertaken extensive investigations into quantifying methane emissions from landfills, utilizing various methodologies beyond reliance on the LandGEM model. Numerous studies have underscored the efficacy of this model in predicting methane generation and exploring its utility in activities such as energy generation and emission mitigation strategies. Objectives: Municipal solid waste landfills are vital contributors to greenhouse gas emissions, notably methane, during waste decomposition. Significant research has focused on estimating methane output from landfills, utilizing various approaches beyond the LandGEM model. Studies highlight the model's potential in assessing methane generation and its relevance to activities like energy generation and emission management. Methods: The LandGEM simulation model is employed to forecast future waste generation and provide valuable insights into the environmental consequences of landfill management on the island. By analyzing population data and waste generation statistics from 2023 to 2042, the study utilizes the LandGEM software to estimate methane emissions, incorporating site-specific data and waste composition analysis. Results: The findings reveal that 36% of the total waste on Kish Island consists of putrescible material, with food waste being a significant component. Over the study period, approximately 1,282,637 tons of waste are projected to be disposed of in the Kish landfill. The LandGEM model predicts the highest landfill gas emission in 2043, primarily methane, along with carbon dioxide and non-metallic organic compounds. The estimated quantities for these gases in 2043 are 5.415E+03, 1.486E+04, and 2.327E+02 metric tons, respectively. Conclusions: This research highlights the importance of tailored waste management strategies to mitigate environmental impacts and emphasizes the need for efficient gas collection systems. A proposed gas collection system combining vertical wells, horizontal trenches, and under-membrane pipes is discussed to enhance efficiency and reduce environmental risks. The study provides valuable insights into waste management and landfill gas emissions on Kish Island, underscoring the significance of accurate emission prediction and effective management to minimize environmental consequences and optimize the utilization of renewable energy resources.

H2O2-Activatable Liposomal Nanobomb Capable of Generating Hypoxia-Irrelevant Alkyl Radicals by Photo-Triggered Cascade Reaction for High-Performance Elimination of Biofilm Bacteria.

High H2O2 levels are widely present at the infection sites or in the biofilm microenvironment. Herein, hemin with peroxidase-like catalytic activity and its substrate, 2,2'-azino-bis(3-ethylbenzothiazoline-6-sulfonic acid) (ABTS), are simultaneously introduced into a liposomal nanoparticle containing thermosensitive 2,2'-azobis[2-(2-imidazolin-2-yl) propane] dihydrochloride (AIBI)-loaded bovine serum albumin (BAG), rationally constructing an H2O2-activatable liposomal nanobomb (Lipo@BHA) for combating biofilm-associated bacterial infections with high performance. In the presence of H2O2, hemin can catalyze the conversion of ABTS into its oxidized form (ABTS·+) with strong near-infrared (NIR) absorption, which produces photonic hyperpyrexia to cause the decomposition of AIBI into oxygen-independent alkyl radicals (·R) and nitrogen (N2) microbubbles. The former not only directly damage bacterial cells but also significantly accelerates the oxidization of ABTS to ABTS·+ for augmenting photothermal-triggered generation of ·R. Interestingly, the released N2 can induce transient cavitation to rupture lysosomal nanoparticle and improve the biofilm permeability, thereby enhancing the antibiofilm effect of Lipo@BHA. The proposed Lipo@BHA exhibits satisfactory multi-mode combination antibacterial properties. Through endogenous H2O2-activated cascade reaction, Lipo@BHA achieves remarkable hypoxia-irrelevant ·R therapy of biofilm-associated wound infections with low cytotoxicity and good in vivo biosafety. Therefore, this work presents a versatile H2O2-activatable cascade ·R generation strategy for biofilm-specific therapeutic applications.

Protein Language Models and Machine Learning Facilitate the Identification of Antimicrobial Peptides.

Peptides are bioactive molecules whose functional versatility in living organisms has led to successful applications in diverse fields. In recent years, the amount of data describing peptide sequences and function collected in open repositories has substantially increased, allowing the application of more complex computational models to study the relations between the peptide composition and function. This work introduces AMP-Detector, a sequence-based classification model for the detection of peptides' functional biological activity, focusing on accelerating the discovery and de novo design of potential antimicrobial peptides (AMPs). AMP-Detector introduces a novel sequence-based pipeline to train binary classification models, integrating protein language models and machine learning algorithms. This pipeline produced 21 models targeting antimicrobial, antiviral, and antibacterial activity, achieving average precision exceeding 83%. Benchmark analyses revealed that our models outperformed existing methods for AMPs and delivered comparable results for other biological activity types. Utilizing the Peptide Atlas, we applied AMP-Detector to discover over 190,000 potential AMPs and demonstrated that it is an integrative approach with generative learning to aid in de novo design, resulting in over 500 novel AMPs. The combination of our methodology, robust models, and a generative design strategy offers a significant advancement in peptide-based drug discovery and represents a pivotal tool for therapeutic applications.

Maternal Immunization Decision-Making Among Pregnant and Lactating People in Kenya: A Qualitative Exploration of Peer Influences on Vaccine Decision-Making for a Future RSV Vaccine

IntroductionRespiratory syncytial virus (RSV) is a leading cause of respiratory illness in infants globally, with new maternal RSV vaccines on the horizon. Vaccine decision-making during pregnancy is shaped by individual, interpersonal, community, and societal factors. This study explored key interpersonal influences on maternal vaccine decision-making among pregnant and lactating people (PLP) and community members in Kenya.MethodsThis qualitative study conducted in-depth interviews with six pregnant people, 18 lactating people, and 10 community members in one rural and one urban county in Kenya. Data were analyzed using a grounded theory approach.ResultsParticipants identified the pregnant person themself, male partners, other family members, peers, and healthcare providers (HCPs) as key influences on the maternal immunization decision-making process. The majority of interviewed PLP believed that decision-making during pregnancy should be left to themselves due to autonomy and their role as the primary caregiver. Community members, including male partners, also identified pregnant people as the key decision-maker. While some PLP said they deferred to male partners to make vaccine decisions, more felt that men were not as informed on maternal and child issues as themselves or other female peers and relatives. HCPs emerged as important influences and information sources for PLP during decision-making.DiscussionUnderstanding who influences vaccine-decision making during pregnancy will help inform demand generation strategies, and in turn, uptake of future maternal vaccines, including RSV vaccines. Given the strong role HCPs and peers have in the decision-making process, targeting key potential influences is essential to improve vaccine acceptance.

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.