针对自动驾驶场景下,近处干扰点云误检率高、远处稀疏点云漏检率高的问题,提出了一种基于改进PointPillars的自动驾驶障碍物点云检测算法.首先,通过聚合模块和共享多层感知机(shared multi-layer perceptron,MLP)对柱体内点云进行特征编...针对自动驾驶场景下,近处干扰点云误检率高、远处稀疏点云漏检率高的问题,提出了一种基于改进PointPillars的自动驾驶障碍物点云检测算法.首先,通过聚合模块和共享多层感知机(shared multi-layer perceptron,MLP)对柱体内点云进行特征编码,采用最大池化与平均池化叠加的方法将点云的显著特征与细节特征映射为柱体特征;其次,针对算法对伪图特征关注与利用不充分的问题,引入坐标注意力(coordinate attention,CA)机制和残差连接的伪图特征提取模块(attention and residual second block,ARSB),将深层与浅层特征图进行融合,优化算法梯度,增强算法对有效目标的关注度.试验结果表明:改进算法对全局点云检测精度较高,平均精度优于PointPillars、稀疏到稠密3D目标检测器(STD)等点云目标检测算法,在汽车类别上的检测精度优势明显,检测速度较快,符合实时性要求.展开更多
Vision Transformers(ViTs)have achieved remarkable success across various artificial intelligence-based computer vision applications.However,their demanding computational and memory requirements pose significant challe...Vision Transformers(ViTs)have achieved remarkable success across various artificial intelligence-based computer vision applications.However,their demanding computational and memory requirements pose significant challenges for de-ployment on resource-constrained edge devices.Although post-training quantization(PTQ)provides a promising solution by reducing model precision with minimal calibration data,aggressive low-bit quantization typically leads to substantial perfor-mance degradation.To address this challenge,we present the truncated uniform-log2 quantizer and progressive bit-decline reconstruction method for vision Transformer quantization(TP-ViT).It is an innovative PTQ framework specifically designed for ViTs,featuring two key technical contributions:(1)truncated uniform-log2 quantizer,a novel quantization approach which effectively handles outlier values in post-Softmax activations,significantly reducing quantization errors;(2)bit-decline optimiza-tion strategy,which employs transition weights to gradually reduce bit precision while maintaining model performance under extreme quantization conditions.Comprehensive experiments on image classification,object detection,and instance segmenta-tion tasks demonstrate TP-ViT’s superior performance compared to state-of-the-art PTQ methods,particularly in challenging 3-bit quantization scenarios.Our framework achieves a notable 6.18 percentage points improvement in top-1 accuracy for ViT-small under 3-bit quantization.These results validate TP-ViT’s robustness and general applicability,paving the way for more efficient deployment of ViT models in computer vision applications on edge hardware.展开更多
Because of the developed surface of the Triply PeriodicMinimumSurface(TPMS)structures,polylactide(PLA)products with a TPMS structure are thought to be promising bio soluble implants with the potential for targeted dru...Because of the developed surface of the Triply PeriodicMinimumSurface(TPMS)structures,polylactide(PLA)products with a TPMS structure are thought to be promising bio soluble implants with the potential for targeted drug delivery.For implants,mechanical properties are key performance characteristics,so understanding the deformation and failure mechanisms is essential for selecting the appropriate implant structure.The deformation and fracture processes in PLA samples with different interior architectures have been studied through computer simulation and experimental research.Two TPMS topologies,the Schwarz Diamond and Gyroid architectures,were used for the sample construction by 3D printing.ANSYS software was utilized to simulate compressive deformation.It was found that under the same load,the vonMises stresses in the Gyroid structure are higher than those in the Schwartz Diamond structure,which was associated with the different orientations of the cells in the studied structures in relation to the direction of the loading axis.The deformation process occurs in the local regions of the studied TPMS structures.Maximum von Mises stresses were observed in the vertical parts of the structures oriented along the load direction.It was found that,unlike the Gyroid,the Schwartz Diamond structure contains a frame that forms unique stiffening ribs,which ensures the redistribution of the load under the vertical loading direction.An analysis of the mechanical characteristics of PLA samples with the Schwartz Diamond and Gyroid structures produced by the Fused Deposition Modeling(FDM)method was correlated with computer simulation.The Schwarz Diamond-type structure was shown to have a higher absorption energy than the Gyroid one.A study of the fracture in PLA samples with various cell sizes revealed a particular feature related to the samples’periodic surface topology and the 3D printing process.Scanning electron microscopic(SEM)studies of the samples deformed by compression showed thatwith an increase in the density of the samples,the failure mechanism changes from ductile to quasi-brittle due to the complex participation of both cell deformation and fiber deformation.展开更多
Large language models(LLMs)have demonstrated significant capabilities in semantic understanding and code generation.However,cybersecurity tasks often require prompting the adaptation of open-source models to this doma...Large language models(LLMs)have demonstrated significant capabilities in semantic understanding and code generation.However,cybersecurity tasks often require prompting the adaptation of open-source models to this domain.Despite their effectiveness,large-parameter LLMs incur substantial memory usage and runtime costs during task inference and downstreamfine-tuning for cybersecurity applications.In this study,we fine-tuned six LLMs with parameters under 4 billion using LoRA(Low-Rank Adaptation)on specific cybersecurity instruction datasets,employing evaluation metrics similar to Hackmentor.Results indicate that post-fine-tuning,smaller models achieved victory or parity rates up to 85%against larger models like Qwen-1.5-14B on cybersecurity test datasets,with the best model reaching a 90%win or tie rate compared to SecGPT.Additionally,these smaller models required significantly less computational resources,reducing fine-tuning times by up to 53%and enhancing efficiency in downstream tasks.Further validation showed that withminimal fine-tuning,our models achieved a performance gain of 21.66%to 31.32%in tactical extraction and 30.69%to 40.42%in technical extraction tasks,significantly outperforming ChatGPT.These findings highlight the potential of smaller parameter LLMs for optimizing performance and resource utilization in cybersecurity applications including methods such as technique and tactic extraction.It will facilitate future research on the application of small-parameter large language models in the cybersecurity domain.展开更多
Triply periodic minimal surface(TPMS)structures are widely utilized in engineering and biomedical fields owing to their superior mechanical and functional properties.However,limited by the current additive manufacturi...Triply periodic minimal surface(TPMS)structures are widely utilized in engineering and biomedical fields owing to their superior mechanical and functional properties.However,limited by the current additive manufacturing(AM)techniques,insufficient wall thickness often leads to poor forming quality or even printing failure.Therefore,accurate prediction of wall thickness parameters during the design stage is essential.This study proposes a prediction approach for the wall thickness parameters of TPMS models by integrating a Convolutional Neural Network–Support Vector Regression(CNN-SVM)framework with Multiple Linear Regression(MLR).A total of 152 TPMS models were randomly generated,resulting in 912 sets of sample data.Voxel-based sampling and rasterization preprocessing were employed to prepare the data for model input.The CNN-SVM model was developed using TPMS type,lattice filling type,volume fraction,and cell length as input features,with wall thickness as the output variable.Subsequently,the MLR method was applied to quantify the influence weights of these parameters.Experimental results demonstrate that the CNN-SVM model achieves a mean squared error(MSE)of 0.0011 and a coefficient of determination(R2)of 0.92.Approximately 86.9%of the test samples exhibited prediction errors within 20%,representing performance improvements of 15.8%,10.6%,and 18.5%over traditional MLR,CNN,and SVM models,respectively.The MLR analysis further indicates that the Sheet filling type exerts the most significant positive effect on wall thickness(0.45729),whereas theDiamond TPMS structure shows the most prominent negative impact(−0.23494).The proposed hybrid model provides an effective and reliable strategy for predicting wall thickness parameters in TPMS-based additive manufacturing designs.展开更多
This paper proposes a robust control-oriented identification method for errors-in-variables(EIV)systems in output feedbacks using frequency-response(FR)experimental data.An important relation between such a closed-loo...This paper proposes a robust control-oriented identification method for errors-in-variables(EIV)systems in output feedbacks using frequency-response(FR)experimental data.An important relation between such a closed-loop EIV system and its coprime factor(CF)uncertainty description is first derived,based on which the FR measurements suitable for plant CF identification are able to be generated.Different factorizations of a given controller in the closed-loop system can be made best use to adjust right coprime factors(RCFs)of the plant so as to realize an improvement on the signal-to-noise ratio of identification experimental data.Subsequently,a nominal RCF model is estimated by linear matrix inequalities from the applicable FR measurements and its associated worst-case errors are quantified from a priori and a posteriori information on the underlying system.A resulting RCF perturbation model set can then be described by the nominal RCF model and its worst-case error bounds.Such a model set capable of being stabilized by the given controller is ready for its robust stabilizing controller redesign and robust performance analysis.Finally,a numerical simulation is given to show the efficacy of the proposed identification method.展开更多
Early detection and treatment of colorectal cancer can effectively reduce the harm caused to patients.However,early symptoms are not obvious.Exosomes are extracellular vesicles that carry information between cells cir...Early detection and treatment of colorectal cancer can effectively reduce the harm caused to patients.However,early symptoms are not obvious.Exosomes are extracellular vesicles that carry information between cells circulating in the human body.Despite the critical importance of tumor-derived exosomes for early cancer detection,prognosis,and treatment guidance,sensitive detection of exosomes remains challenging.We report a novel immunosensor for the detection of exosomes using graphene oxide(GO)/Au/Ag as the electrochemical sensing platform and spherical Au@PtPd porous nanoparticles to provide signal amplification.A large-surface-area GO/Au/Ag-modified screen-printed carbon electrode(SPCE)is used to immobilize CD63 aptamers.The prepared strawberry-like Au@PtPd demonstrates improved sensitivity and electrocatalytic activity with increasing aptamer load.Meanwhile,its good conductivity accelerates the electron transfer rate on the SPCE.The immunosensor exhibits a linear range(100–1.0×10^(6)exosomes·μl^(−1))and a detection limit of 23 exosomes·μl^(−1).Furthermore,its potential for the detection of exosomes provides a new clinical application for early diagnosis of colorectal tumors.展开更多
This paper proposes a deep learning-based 3D LiDAR perception framework designed for applications such as autonomous robots and vehicles.To address the high dependency on large-scale annotated data—an inherent limita...This paper proposes a deep learning-based 3D LiDAR perception framework designed for applications such as autonomous robots and vehicles.To address the high dependency on large-scale annotated data—an inherent limitation of deep learning models—this study introduces a hybrid perception architecture that incorporates expertdriven LiDAR processing techniques into the deep neural network.Traditional 3DLiDAR processingmethods typically remove ground planes and apply distance-or density-based clustering for object detection.In this work,such expert knowledge is encoded as feature-level inputs and fused with the deep network,therebymitigating the data dependency issue of conventional learning-based approaches.Specifically,the proposedmethod combines two expert algorithms—Patchwork++for ground segmentation and DBSCAN for clustering—with a PointPillars-based LiDAR detection network.We design four hybrid versions of the network depending on the stage and method of integrating expert features into the feature map of the deep model.Among these,Version 4 incorporates a modified neck structure in PointPillars and introduces a new Cluster 2D Pseudo-Map Branch that utilizes cluster-level pseudo-images generated from Patchwork++and DBSCAN.This version achieved a+3.88%improvement mean Average Precision(mAP)compared to the baseline PointPillars.The results demonstrate that embedding expert-based perception logic into deep neural architectures can effectively enhance performance and reduce dependency on extensive training datasets,offering a promising direction for robust 3D LiDAR object detection in real-world scenarios.展开更多
To enhance the accuracy of short-term photovoltaic power output prediction and address issues such as insufficient spatial resolution of meteorological forecast data and weak generalization ability of models,this pape...To enhance the accuracy of short-term photovoltaic power output prediction and address issues such as insufficient spatial resolution of meteorological forecast data and weak generalization ability of models,this paper proposes a prediction method that integrates spatial downscaling meteorological data with a convolutional neural network(CNN)-iTransformer-long short-term memory(LSTM)model.First,the rime-optimized random forest regression algorithm(RIME-RF)is employed to perform spatial downscaling on numerical weather prediction(NWP)data,thereby improving its local applicability.Second,a CNN-iTransformer-LSTM hybrid prediction model is constructed.This model utilizes a CNN as a spatial feature extractor to capture local patterns in meteorological data,employs an iTransformer to model the global dependencies among multiple variables,and leverages an LSTM to enhance the learning of short-term temporal dynamic features,thereby achieving efficient collaborative mining of multi-scale features.Finally,experiments are conducted using actual data from a photovoltaic power station in Hebei,China,during various seasons and weather conditions.The results show that the proposed model outperforms the comparison models in terms of the root mean square error(RMSE),mean absolute error(MAE),and R2,maintaining high prediction accuracy and stability even under complex weather conditions such as overcast and rainy days.The downscaling process further enhances the prediction performance,verifying the effectiveness and practicality of this method.展开更多
文摘针对自动驾驶场景下,近处干扰点云误检率高、远处稀疏点云漏检率高的问题,提出了一种基于改进PointPillars的自动驾驶障碍物点云检测算法.首先,通过聚合模块和共享多层感知机(shared multi-layer perceptron,MLP)对柱体内点云进行特征编码,采用最大池化与平均池化叠加的方法将点云的显著特征与细节特征映射为柱体特征;其次,针对算法对伪图特征关注与利用不充分的问题,引入坐标注意力(coordinate attention,CA)机制和残差连接的伪图特征提取模块(attention and residual second block,ARSB),将深层与浅层特征图进行融合,优化算法梯度,增强算法对有效目标的关注度.试验结果表明:改进算法对全局点云检测精度较高,平均精度优于PointPillars、稀疏到稠密3D目标检测器(STD)等点云目标检测算法,在汽车类别上的检测精度优势明显,检测速度较快,符合实时性要求.
基金supported by the National Natural Science Foundation of China(Nos.62301092 and 62301093).
文摘Vision Transformers(ViTs)have achieved remarkable success across various artificial intelligence-based computer vision applications.However,their demanding computational and memory requirements pose significant challenges for de-ployment on resource-constrained edge devices.Although post-training quantization(PTQ)provides a promising solution by reducing model precision with minimal calibration data,aggressive low-bit quantization typically leads to substantial perfor-mance degradation.To address this challenge,we present the truncated uniform-log2 quantizer and progressive bit-decline reconstruction method for vision Transformer quantization(TP-ViT).It is an innovative PTQ framework specifically designed for ViTs,featuring two key technical contributions:(1)truncated uniform-log2 quantizer,a novel quantization approach which effectively handles outlier values in post-Softmax activations,significantly reducing quantization errors;(2)bit-decline optimiza-tion strategy,which employs transition weights to gradually reduce bit precision while maintaining model performance under extreme quantization conditions.Comprehensive experiments on image classification,object detection,and instance segmenta-tion tasks demonstrate TP-ViT’s superior performance compared to state-of-the-art PTQ methods,particularly in challenging 3-bit quantization scenarios.Our framework achieves a notable 6.18 percentage points improvement in top-1 accuracy for ViT-small under 3-bit quantization.These results validate TP-ViT’s robustness and general applicability,paving the way for more efficient deployment of ViT models in computer vision applications on edge hardware.
文摘Because of the developed surface of the Triply PeriodicMinimumSurface(TPMS)structures,polylactide(PLA)products with a TPMS structure are thought to be promising bio soluble implants with the potential for targeted drug delivery.For implants,mechanical properties are key performance characteristics,so understanding the deformation and failure mechanisms is essential for selecting the appropriate implant structure.The deformation and fracture processes in PLA samples with different interior architectures have been studied through computer simulation and experimental research.Two TPMS topologies,the Schwarz Diamond and Gyroid architectures,were used for the sample construction by 3D printing.ANSYS software was utilized to simulate compressive deformation.It was found that under the same load,the vonMises stresses in the Gyroid structure are higher than those in the Schwartz Diamond structure,which was associated with the different orientations of the cells in the studied structures in relation to the direction of the loading axis.The deformation process occurs in the local regions of the studied TPMS structures.Maximum von Mises stresses were observed in the vertical parts of the structures oriented along the load direction.It was found that,unlike the Gyroid,the Schwartz Diamond structure contains a frame that forms unique stiffening ribs,which ensures the redistribution of the load under the vertical loading direction.An analysis of the mechanical characteristics of PLA samples with the Schwartz Diamond and Gyroid structures produced by the Fused Deposition Modeling(FDM)method was correlated with computer simulation.The Schwarz Diamond-type structure was shown to have a higher absorption energy than the Gyroid one.A study of the fracture in PLA samples with various cell sizes revealed a particular feature related to the samples’periodic surface topology and the 3D printing process.Scanning electron microscopic(SEM)studies of the samples deformed by compression showed thatwith an increase in the density of the samples,the failure mechanism changes from ductile to quasi-brittle due to the complex participation of both cell deformation and fiber deformation.
文摘Large language models(LLMs)have demonstrated significant capabilities in semantic understanding and code generation.However,cybersecurity tasks often require prompting the adaptation of open-source models to this domain.Despite their effectiveness,large-parameter LLMs incur substantial memory usage and runtime costs during task inference and downstreamfine-tuning for cybersecurity applications.In this study,we fine-tuned six LLMs with parameters under 4 billion using LoRA(Low-Rank Adaptation)on specific cybersecurity instruction datasets,employing evaluation metrics similar to Hackmentor.Results indicate that post-fine-tuning,smaller models achieved victory or parity rates up to 85%against larger models like Qwen-1.5-14B on cybersecurity test datasets,with the best model reaching a 90%win or tie rate compared to SecGPT.Additionally,these smaller models required significantly less computational resources,reducing fine-tuning times by up to 53%and enhancing efficiency in downstream tasks.Further validation showed that withminimal fine-tuning,our models achieved a performance gain of 21.66%to 31.32%in tactical extraction and 30.69%to 40.42%in technical extraction tasks,significantly outperforming ChatGPT.These findings highlight the potential of smaller parameter LLMs for optimizing performance and resource utilization in cybersecurity applications including methods such as technique and tactic extraction.It will facilitate future research on the application of small-parameter large language models in the cybersecurity domain.
基金funded by Sichuan Provincial Regional Innovation Cooperation Project Funding,grant number 2024YFHZ0073Funded by the Research Innovation Team Programof Sichuan University of Science and Chemical Technology,grant number SUSE652A015.
文摘Triply periodic minimal surface(TPMS)structures are widely utilized in engineering and biomedical fields owing to their superior mechanical and functional properties.However,limited by the current additive manufacturing(AM)techniques,insufficient wall thickness often leads to poor forming quality or even printing failure.Therefore,accurate prediction of wall thickness parameters during the design stage is essential.This study proposes a prediction approach for the wall thickness parameters of TPMS models by integrating a Convolutional Neural Network–Support Vector Regression(CNN-SVM)framework with Multiple Linear Regression(MLR).A total of 152 TPMS models were randomly generated,resulting in 912 sets of sample data.Voxel-based sampling and rasterization preprocessing were employed to prepare the data for model input.The CNN-SVM model was developed using TPMS type,lattice filling type,volume fraction,and cell length as input features,with wall thickness as the output variable.Subsequently,the MLR method was applied to quantify the influence weights of these parameters.Experimental results demonstrate that the CNN-SVM model achieves a mean squared error(MSE)of 0.0011 and a coefficient of determination(R2)of 0.92.Approximately 86.9%of the test samples exhibited prediction errors within 20%,representing performance improvements of 15.8%,10.6%,and 18.5%over traditional MLR,CNN,and SVM models,respectively.The MLR analysis further indicates that the Sheet filling type exerts the most significant positive effect on wall thickness(0.45729),whereas theDiamond TPMS structure shows the most prominent negative impact(−0.23494).The proposed hybrid model provides an effective and reliable strategy for predicting wall thickness parameters in TPMS-based additive manufacturing designs.
文摘This paper proposes a robust control-oriented identification method for errors-in-variables(EIV)systems in output feedbacks using frequency-response(FR)experimental data.An important relation between such a closed-loop EIV system and its coprime factor(CF)uncertainty description is first derived,based on which the FR measurements suitable for plant CF identification are able to be generated.Different factorizations of a given controller in the closed-loop system can be made best use to adjust right coprime factors(RCFs)of the plant so as to realize an improvement on the signal-to-noise ratio of identification experimental data.Subsequently,a nominal RCF model is estimated by linear matrix inequalities from the applicable FR measurements and its associated worst-case errors are quantified from a priori and a posteriori information on the underlying system.A resulting RCF perturbation model set can then be described by the nominal RCF model and its worst-case error bounds.Such a model set capable of being stabilized by the given controller is ready for its robust stabilizing controller redesign and robust performance analysis.Finally,a numerical simulation is given to show the efficacy of the proposed identification method.
基金supported by National Natural Science Foundation of China(Grant No.12105239)the Science and Technology Projects of Henan Province(Grant No.232102310002).
文摘Early detection and treatment of colorectal cancer can effectively reduce the harm caused to patients.However,early symptoms are not obvious.Exosomes are extracellular vesicles that carry information between cells circulating in the human body.Despite the critical importance of tumor-derived exosomes for early cancer detection,prognosis,and treatment guidance,sensitive detection of exosomes remains challenging.We report a novel immunosensor for the detection of exosomes using graphene oxide(GO)/Au/Ag as the electrochemical sensing platform and spherical Au@PtPd porous nanoparticles to provide signal amplification.A large-surface-area GO/Au/Ag-modified screen-printed carbon electrode(SPCE)is used to immobilize CD63 aptamers.The prepared strawberry-like Au@PtPd demonstrates improved sensitivity and electrocatalytic activity with increasing aptamer load.Meanwhile,its good conductivity accelerates the electron transfer rate on the SPCE.The immunosensor exhibits a linear range(100–1.0×10^(6)exosomes·μl^(−1))and a detection limit of 23 exosomes·μl^(−1).Furthermore,its potential for the detection of exosomes provides a new clinical application for early diagnosis of colorectal tumors.
基金supported by Basic Science Research Program through the National Research Foundation of Korea(NRF)funded by the Ministry of Education(RS-2023-00245084)by Korea Institute for Advancement of Technology(KIAT)grant funded by the Korea Government(MOTIE)(RS-2024-00415938,HRD Program for Industrial Innovation)and Soonchunhyang University.
文摘This paper proposes a deep learning-based 3D LiDAR perception framework designed for applications such as autonomous robots and vehicles.To address the high dependency on large-scale annotated data—an inherent limitation of deep learning models—this study introduces a hybrid perception architecture that incorporates expertdriven LiDAR processing techniques into the deep neural network.Traditional 3DLiDAR processingmethods typically remove ground planes and apply distance-or density-based clustering for object detection.In this work,such expert knowledge is encoded as feature-level inputs and fused with the deep network,therebymitigating the data dependency issue of conventional learning-based approaches.Specifically,the proposedmethod combines two expert algorithms—Patchwork++for ground segmentation and DBSCAN for clustering—with a PointPillars-based LiDAR detection network.We design four hybrid versions of the network depending on the stage and method of integrating expert features into the feature map of the deep model.Among these,Version 4 incorporates a modified neck structure in PointPillars and introduces a new Cluster 2D Pseudo-Map Branch that utilizes cluster-level pseudo-images generated from Patchwork++and DBSCAN.This version achieved a+3.88%improvement mean Average Precision(mAP)compared to the baseline PointPillars.The results demonstrate that embedding expert-based perception logic into deep neural architectures can effectively enhance performance and reduce dependency on extensive training datasets,offering a promising direction for robust 3D LiDAR object detection in real-world scenarios.
文摘To enhance the accuracy of short-term photovoltaic power output prediction and address issues such as insufficient spatial resolution of meteorological forecast data and weak generalization ability of models,this paper proposes a prediction method that integrates spatial downscaling meteorological data with a convolutional neural network(CNN)-iTransformer-long short-term memory(LSTM)model.First,the rime-optimized random forest regression algorithm(RIME-RF)is employed to perform spatial downscaling on numerical weather prediction(NWP)data,thereby improving its local applicability.Second,a CNN-iTransformer-LSTM hybrid prediction model is constructed.This model utilizes a CNN as a spatial feature extractor to capture local patterns in meteorological data,employs an iTransformer to model the global dependencies among multiple variables,and leverages an LSTM to enhance the learning of short-term temporal dynamic features,thereby achieving efficient collaborative mining of multi-scale features.Finally,experiments are conducted using actual data from a photovoltaic power station in Hebei,China,during various seasons and weather conditions.The results show that the proposed model outperforms the comparison models in terms of the root mean square error(RMSE),mean absolute error(MAE),and R2,maintaining high prediction accuracy and stability even under complex weather conditions such as overcast and rainy days.The downscaling process further enhances the prediction performance,verifying the effectiveness and practicality of this method.
