The von Neumann bottleneck in conventional computing architectures presents a significant challenge for data-inten-sive artificial intelligence applications.A promising approach involves designing specialized hardware...The von Neumann bottleneck in conventional computing architectures presents a significant challenge for data-inten-sive artificial intelligence applications.A promising approach involves designing specialized hardware with on-chip parameter tunability,which directly accelerates machine learning functions.This work demonstrates a continuously tunable mixed-kernel function physically realized within a van der Waals heterostructure.We designed and fabricated a MoTe_(2)/MoS_(2)type-Ⅱvertical heterojunction phototransistor,which exhibits a non-monotonic,Gaussian-like optoelectronic response owing to its unique inter-layer charge transfer mechanism.This intrinsic physical behavior directly maps to a mixed-kernel function combining Gaussian and Sigmoid characteristics.Furthermore,the hardware kernel can be continuously modulated by in-situ tuning of external opti-cal stimuli.The mixed-kernel exhibited exceptional performance,achieving precision,accuracy,and area under the curve(AUC)values of 95.8%,96%,and 0.9986,respectively,significantly outperforming conventional kernels.By successfully embedding a complex,adaptable mathematical function into the intrinsic physical properties of a single device,this work pioneers a novel pathway toward next-generation,energy-efficient intelligent systems with hardware-level adaptability.展开更多
Modern intelligent systems,such as autonomous vehicles and face recognition,must continuously adapt to new scenarios while preserving their ability to handle previously encountered situations.However,when neural netwo...Modern intelligent systems,such as autonomous vehicles and face recognition,must continuously adapt to new scenarios while preserving their ability to handle previously encountered situations.However,when neural networks learn new classes sequentially,they suffer from catastrophic forgetting—the tendency to lose knowledge of earlier classes.This challenge,which lies at the core of class-incremental learning,severely limits the deployment of continual learning systems in real-world applications with streaming data.Existing approaches,including rehearsalbased methods and knowledge distillation techniques,have attempted to address this issue but often struggle to effectively preserve decision boundaries and discriminative features under limited memory constraints.To overcome these limitations,we propose a support vector-guided framework for class-incremental learning.The framework integrates an enhanced feature extractor with a Support Vector Machine classifier,which generates boundary-critical support vectors to guide both replay and distillation.Building on this architecture,we design a joint feature retention strategy that combines boundary proximity with feature diversity,and a Support Vector Distillation Loss that enforces dual alignment in decision and semantic spaces.In addition,triple attention modules are incorporated into the feature extractor to enhance representation power.Extensive experiments on CIFAR-100 and Tiny-ImageNet demonstrate effective improvements.On CIFAR-100 and Tiny-ImageNet with 5 tasks,our method achieves 71.68%and 58.61%average accuracy,outperforming strong baselines by 3.34%and 2.05%.These advantages are consistently observed across different task splits,highlighting the robustness and generalization of the proposed approach.Beyond benchmark evaluations,the framework also shows potential in few-shot and resource-constrained applications such as edge computing and mobile robotics.展开更多
The support vector machine,a widely used binary classification method,may expose sensitive information during training.To address this,the authors propose a personalized differential privacy method that extends differ...The support vector machine,a widely used binary classification method,may expose sensitive information during training.To address this,the authors propose a personalized differential privacy method that extends differential privacy.Specifically,the authors introduce personalized differentially private support vector machines to meet different individuals'privacy requirements,using a reweighting strategy and the Laplace mechanism.Theoretical analysis demonstrates that the proposed methods simultaneously satisfy the requirements of personalized differential privacy and ensure model prediction accuracy at these privacy levels.Extensive experiments demonstrate that the proposed methods outperform the existing methods.展开更多
The development of clinical candidates that modify the natural progression of sporadic Parkinson's disease and related synucleinopathies is a praiseworthy endeavor,but extremely challenging.Therapeutic candidates ...The development of clinical candidates that modify the natural progression of sporadic Parkinson's disease and related synucleinopathies is a praiseworthy endeavor,but extremely challenging.Therapeutic candidates that were successful in preclinical Parkinson's disease animal models have repeatedly failed when tested in clinical trials.While these failures have many possible explanations,it is perhaps time to recognize that the problem lies with the animal models rather than the putative candidate.In other words,the lack of adequate animal models of Parkinson's disease currently represents the main barrier to preclinical identification of potential disease-modifying therapies likely to succeed in clinical trials.However,this barrier may be overcome by the recent introduction of novel generations of viral vectors coding for different forms of alpha-synuclein species and related genes.Although still facing several limitations,these models have managed to mimic the known neuropathological hallmarks of Parkinson's disease with unprecedented accuracy,delineating a more optimistic scenario for the near future.展开更多
Fluidic Thrust Vectoring(FTV)is used for the yaw attitude control of tailless flying wing,which can significantly improve stealth performance,maneuverability and lateral/heading maneuverability.The FTV control scheme ...Fluidic Thrust Vectoring(FTV)is used for the yaw attitude control of tailless flying wing,which can significantly improve stealth performance,maneuverability and lateral/heading maneuverability.The FTV control scheme of co-directional secondary flow was designed based on a 30 kgf thrust turbojet engine,an equivalent rudder deflection control variable of Mass Flow Combination(MFC)was proposed,and a control model was established to form a FTV control system scheme,which was integrated with the flight control system of a 100 kg tailless flying wing with medium aspect ratio to achieve closed-loop control of the yaw attitude based on FTV.The heading stability augmentation and maneuvering control characteristics and time response characteristics of tailless flying wing by FTV were quantitatively studied through virtual flight test in a wind tunnel at a wind speed of 35 m/s.The results show that the control strategy based on MFC achieves bidirectional continuous and stable control of thrust vector angle in a range of±11°,and the thrust vector angle varies monotonically with MFC;the co-directional FTV realizes bidirectional continuous and stable control of the yaw attitude of tailless flying wing,without longitudinal/lateral coupling moment.The increment of the maximum yawing moment coefficient is 0.0029,the maximum yaw rate is 7.55(°)/s,and the response time of the yaw rate of the vectoring nozzle actuated by the secondary flow is about 0.06 s,which satisfies the heading stability augmentation and maneuvering control response requirements of the aircraft with statically unstable heading,and provides new control means for the heading rudderless attitude control of tailless flying wing.展开更多
The advantages of genome selection(GS) in animal and plant breeding are self-evident.Traditional parametric models have disadvantage in better fit the increasingly large sequencing data and capture complex effects acc...The advantages of genome selection(GS) in animal and plant breeding are self-evident.Traditional parametric models have disadvantage in better fit the increasingly large sequencing data and capture complex effects accurately.Machine learning models have demonstrated remarkable potential in addressing these challenges.In this study,we introduced the concept of mixed kernel functions to explore the performance of support vector machine regression(SVR) in GS.Six single kernel functions(SVR_L,SVR_C,SVR_G,SVR_P,SVR_S,SVR_L) and four mixed kernel functions(SVR_GS,SVR_GP,SVR_LS,SVR_LP) were used to predict genome breeding values.The prediction accuracy,mean squared error(MSE) and mean absolute error(MAE) were used as evaluation indicators to compare with two traditional parametric models(GBLUP,BayesB) and two popular machine learning models(RF,KcRR).The results indicate that in most cases,the performance of the mixed kernel function model significantly outperforms that of GBLUP,BayesB and single kernel function.For instance,for T1 in the pig dataset,the predictive accuracy of SVR_GS is improved by 10% compared to GBLUP,and by approximately 4.4 and 18.6% compared to SVR_G and SVR_S respectively.For E1 in the wheat dataset,SVR_GS achieves 13.3% higher prediction accuracy than GBLUP.Among single kernel functions,the Laplacian and Gaussian kernel functions yield similar results,with the Gaussian kernel function performing better.The mixed kernel function notably reduces the MSE and MAE when compared to all single kernel functions.Furthermore,regarding runtime,SVR_GS and SVR_GP mixed kernel functions run approximately three times faster than GBLUP in the pig dataset,with only a slight increase in runtime compared to the single kernel function model.In summary,the mixed kernel function model of SVR demonstrates speed and accuracy competitiveness,and the model such as SVR_GS has important application potential for GS.展开更多
In position-sensorless brushless direct current(DC)motors(BLDCMs)fed by a four-switch three-phase(FSTP)inverter,only two phases are fully controlled,while the remaining phase is tied to the midpoint of the split DC-li...In position-sensorless brushless direct current(DC)motors(BLDCMs)fed by a four-switch three-phase(FSTP)inverter,only two phases are fully controlled,while the remaining phase is tied to the midpoint of the split DC-link capacitors.The voltage pulses required by inductance-based initial position detection can cause unequal discharge of the series capacitors,shifting the neutral-point voltage away from half of DC-link voltage(U_(dc)/2).This neutral-point drift breaks the spatial symmetry of the inverter voltage vectors,so the 360°electrical period can no longer be evenly partitioned into six sectors during initial rotor position detection.To address this issue,this paper proposes a detection-pulse injection sequence that explicitly accounts for the asymmetric voltage vectors of the FSTP inverter.With the proposed sequence,the initial rotor position can be identified within a 30°electrical sector.The method requires no additional voltage or current sensors,and experimental results confirm its feasibility.展开更多
Parkinson’s disease(PD)is a debilitating neurological disorder affecting over 10 million people worldwide.PD classification models using voice signals as input are common in the literature.It is believed that using d...Parkinson’s disease(PD)is a debilitating neurological disorder affecting over 10 million people worldwide.PD classification models using voice signals as input are common in the literature.It is believed that using deep learning algorithms further enhances performance;nevertheless,it is challenging due to the nature of small-scale and imbalanced PD datasets.This paper proposed a convolutional neural network-based deep support vector machine(CNN-DSVM)to automate the feature extraction process using CNN and extend the conventional SVM to a DSVM for better classification performance in small-scale PD datasets.A customized kernel function reduces the impact of biased classification towards the majority class(healthy candidates in our consideration).An improved generative adversarial network(IGAN)was designed to generate additional training data to enhance the model’s performance.For performance evaluation,the proposed algorithm achieves a sensitivity of 97.6%and a specificity of 97.3%.The performance comparison is evaluated from five perspectives,including comparisons with different data generation algorithms,feature extraction techniques,kernel functions,and existing works.Results reveal the effectiveness of the IGAN algorithm,which improves the sensitivity and specificity by 4.05%–4.72%and 4.96%–5.86%,respectively;and the effectiveness of the CNN-DSVM algorithm,which improves the sensitivity by 1.24%–57.4%and specificity by 1.04%–163%and reduces biased detection towards the majority class.The ablation experiments confirm the effectiveness of individual components.Two future research directions have also been suggested.展开更多
目的:探索通用转录因子ⅡE肽1(GTF2E1)在结直肠癌(CRC)中的表达水平及其对肿瘤细胞增殖的影响,以评估其作为潜在生物标志物和治疗靶点的价值。方法:通过收集并分析来自公共数据库的CRC样本数据,结合临床样本的免疫组化结果,利用单细胞...目的:探索通用转录因子ⅡE肽1(GTF2E1)在结直肠癌(CRC)中的表达水平及其对肿瘤细胞增殖的影响,以评估其作为潜在生物标志物和治疗靶点的价值。方法:通过收集并分析来自公共数据库的CRC样本数据,结合临床样本的免疫组化结果,利用单细胞测序技术和成簇规律间隔短回文重复序列(CRISPR)敲除实验,全面评价GTF2E1在CRC中的表达模式和作用。结果:GTF2E1在CRC组织中呈显著高表达,公共数据集、单细胞分析及内部临床样本检测结果一致。其诊断CRC的综合受试者工作特征曲线下面积为0.87,具有良好的诊断效能;CRISPR敲除实验显示,GTF 2 E 1敲除可显著抑制61种CRC细胞系的增殖。结论:GTF2E1在CRC的发生与增殖中起重要作用,有潜力成为CRC的生物标志物和治疗靶点。展开更多
基金co-supported by the National Natural Science Foundation of China(Grant Nos.62222404,T2450054,62304084,62504087,62361136587 and 92248304)the National Key Research and Development Plan of China(Grant No.2021YFB3601200)+3 种基金the Major Program of Hubei Province(Grant No.2023BAA009)the Research Grants Council of Hong Kong Postdoctoral Fellowship Scheme(Grant No.PDFS2223-4S06)the China Postdoctoral Science Foundation funded project(Grant No.2025M770530)the Postdoctoral Fellowship Program of CPSF(Grant No.GZB20250136).
文摘The von Neumann bottleneck in conventional computing architectures presents a significant challenge for data-inten-sive artificial intelligence applications.A promising approach involves designing specialized hardware with on-chip parameter tunability,which directly accelerates machine learning functions.This work demonstrates a continuously tunable mixed-kernel function physically realized within a van der Waals heterostructure.We designed and fabricated a MoTe_(2)/MoS_(2)type-Ⅱvertical heterojunction phototransistor,which exhibits a non-monotonic,Gaussian-like optoelectronic response owing to its unique inter-layer charge transfer mechanism.This intrinsic physical behavior directly maps to a mixed-kernel function combining Gaussian and Sigmoid characteristics.Furthermore,the hardware kernel can be continuously modulated by in-situ tuning of external opti-cal stimuli.The mixed-kernel exhibited exceptional performance,achieving precision,accuracy,and area under the curve(AUC)values of 95.8%,96%,and 0.9986,respectively,significantly outperforming conventional kernels.By successfully embedding a complex,adaptable mathematical function into the intrinsic physical properties of a single device,this work pioneers a novel pathway toward next-generation,energy-efficient intelligent systems with hardware-level adaptability.
基金supported by the Gansu Provincial Natural Science Foundation(grant number 25JRRA074)the Gansu Provincial Key R&D Science and Technology Program(grant number 24YFGA060)the National Natural Science Foundation of China(grant number 62161019).
文摘Modern intelligent systems,such as autonomous vehicles and face recognition,must continuously adapt to new scenarios while preserving their ability to handle previously encountered situations.However,when neural networks learn new classes sequentially,they suffer from catastrophic forgetting—the tendency to lose knowledge of earlier classes.This challenge,which lies at the core of class-incremental learning,severely limits the deployment of continual learning systems in real-world applications with streaming data.Existing approaches,including rehearsalbased methods and knowledge distillation techniques,have attempted to address this issue but often struggle to effectively preserve decision boundaries and discriminative features under limited memory constraints.To overcome these limitations,we propose a support vector-guided framework for class-incremental learning.The framework integrates an enhanced feature extractor with a Support Vector Machine classifier,which generates boundary-critical support vectors to guide both replay and distillation.Building on this architecture,we design a joint feature retention strategy that combines boundary proximity with feature diversity,and a Support Vector Distillation Loss that enforces dual alignment in decision and semantic spaces.In addition,triple attention modules are incorporated into the feature extractor to enhance representation power.Extensive experiments on CIFAR-100 and Tiny-ImageNet demonstrate effective improvements.On CIFAR-100 and Tiny-ImageNet with 5 tasks,our method achieves 71.68%and 58.61%average accuracy,outperforming strong baselines by 3.34%and 2.05%.These advantages are consistently observed across different task splits,highlighting the robustness and generalization of the proposed approach.Beyond benchmark evaluations,the framework also shows potential in few-shot and resource-constrained applications such as edge computing and mobile robotics.
基金supported by the National Key R&D Program of China under Grant No.2023YFA1008702the National Natural Science Foundation of China under Grant No.12571300。
文摘The support vector machine,a widely used binary classification method,may expose sensitive information during training.To address this,the authors propose a personalized differential privacy method that extends differential privacy.Specifically,the authors introduce personalized differentially private support vector machines to meet different individuals'privacy requirements,using a reweighting strategy and the Laplace mechanism.Theoretical analysis demonstrates that the proposed methods simultaneously satisfy the requirements of personalized differential privacy and ensure model prediction accuracy at these privacy levels.Extensive experiments demonstrate that the proposed methods outperform the existing methods.
基金supported by grants PID2020-120308RB-I00 and PID2023-147802OB-I00 funded by MICIU/AEI/10.13039/501100011033FEDER,UE,by Aligning Science Across Parkinson’s(ref.ASAP-020505)through the Michael J.Fox Foundation for Parkinson’s Research+1 种基金by CiberNed Intramural Collaborative Projects(ref.PI2020/09)by the Spanish Fundación Mutua Madrile?a de Investigación Médica(to JLL)。
文摘The development of clinical candidates that modify the natural progression of sporadic Parkinson's disease and related synucleinopathies is a praiseworthy endeavor,but extremely challenging.Therapeutic candidates that were successful in preclinical Parkinson's disease animal models have repeatedly failed when tested in clinical trials.While these failures have many possible explanations,it is perhaps time to recognize that the problem lies with the animal models rather than the putative candidate.In other words,the lack of adequate animal models of Parkinson's disease currently represents the main barrier to preclinical identification of potential disease-modifying therapies likely to succeed in clinical trials.However,this barrier may be overcome by the recent introduction of novel generations of viral vectors coding for different forms of alpha-synuclein species and related genes.Although still facing several limitations,these models have managed to mimic the known neuropathological hallmarks of Parkinson's disease with unprecedented accuracy,delineating a more optimistic scenario for the near future.
文摘Fluidic Thrust Vectoring(FTV)is used for the yaw attitude control of tailless flying wing,which can significantly improve stealth performance,maneuverability and lateral/heading maneuverability.The FTV control scheme of co-directional secondary flow was designed based on a 30 kgf thrust turbojet engine,an equivalent rudder deflection control variable of Mass Flow Combination(MFC)was proposed,and a control model was established to form a FTV control system scheme,which was integrated with the flight control system of a 100 kg tailless flying wing with medium aspect ratio to achieve closed-loop control of the yaw attitude based on FTV.The heading stability augmentation and maneuvering control characteristics and time response characteristics of tailless flying wing by FTV were quantitatively studied through virtual flight test in a wind tunnel at a wind speed of 35 m/s.The results show that the control strategy based on MFC achieves bidirectional continuous and stable control of thrust vector angle in a range of±11°,and the thrust vector angle varies monotonically with MFC;the co-directional FTV realizes bidirectional continuous and stable control of the yaw attitude of tailless flying wing,without longitudinal/lateral coupling moment.The increment of the maximum yawing moment coefficient is 0.0029,the maximum yaw rate is 7.55(°)/s,and the response time of the yaw rate of the vectoring nozzle actuated by the secondary flow is about 0.06 s,which satisfies the heading stability augmentation and maneuvering control response requirements of the aircraft with statically unstable heading,and provides new control means for the heading rudderless attitude control of tailless flying wing.
基金supported by the China Agriculture Research System of MOF and MARAthe National Natural Science Foundation of China (31872337 and 31501919)the Agricultural Science and Technology Innovation Project,China (ASTIP-IAS02)。
文摘The advantages of genome selection(GS) in animal and plant breeding are self-evident.Traditional parametric models have disadvantage in better fit the increasingly large sequencing data and capture complex effects accurately.Machine learning models have demonstrated remarkable potential in addressing these challenges.In this study,we introduced the concept of mixed kernel functions to explore the performance of support vector machine regression(SVR) in GS.Six single kernel functions(SVR_L,SVR_C,SVR_G,SVR_P,SVR_S,SVR_L) and four mixed kernel functions(SVR_GS,SVR_GP,SVR_LS,SVR_LP) were used to predict genome breeding values.The prediction accuracy,mean squared error(MSE) and mean absolute error(MAE) were used as evaluation indicators to compare with two traditional parametric models(GBLUP,BayesB) and two popular machine learning models(RF,KcRR).The results indicate that in most cases,the performance of the mixed kernel function model significantly outperforms that of GBLUP,BayesB and single kernel function.For instance,for T1 in the pig dataset,the predictive accuracy of SVR_GS is improved by 10% compared to GBLUP,and by approximately 4.4 and 18.6% compared to SVR_G and SVR_S respectively.For E1 in the wheat dataset,SVR_GS achieves 13.3% higher prediction accuracy than GBLUP.Among single kernel functions,the Laplacian and Gaussian kernel functions yield similar results,with the Gaussian kernel function performing better.The mixed kernel function notably reduces the MSE and MAE when compared to all single kernel functions.Furthermore,regarding runtime,SVR_GS and SVR_GP mixed kernel functions run approximately three times faster than GBLUP in the pig dataset,with only a slight increase in runtime compared to the single kernel function model.In summary,the mixed kernel function model of SVR demonstrates speed and accuracy competitiveness,and the model such as SVR_GS has important application potential for GS.
基金supported in part by the National Natural Science Foundation of China under Grant 52477060in part by the Tianjin Natural Science Foundation Project under Grant 24JCZDJC00250in part by the Zhejiang Leading Innovation and Entrepreneurship Team Project under Grant 2024R01012.
文摘In position-sensorless brushless direct current(DC)motors(BLDCMs)fed by a four-switch three-phase(FSTP)inverter,only two phases are fully controlled,while the remaining phase is tied to the midpoint of the split DC-link capacitors.The voltage pulses required by inductance-based initial position detection can cause unequal discharge of the series capacitors,shifting the neutral-point voltage away from half of DC-link voltage(U_(dc)/2).This neutral-point drift breaks the spatial symmetry of the inverter voltage vectors,so the 360°electrical period can no longer be evenly partitioned into six sectors during initial rotor position detection.To address this issue,this paper proposes a detection-pulse injection sequence that explicitly accounts for the asymmetric voltage vectors of the FSTP inverter.With the proposed sequence,the initial rotor position can be identified within a 30°electrical sector.The method requires no additional voltage or current sensors,and experimental results confirm its feasibility.
基金The work described in this paper was fully supported by a grant from Hong Kong Metropolitan University(RIF/2021/05).
文摘Parkinson’s disease(PD)is a debilitating neurological disorder affecting over 10 million people worldwide.PD classification models using voice signals as input are common in the literature.It is believed that using deep learning algorithms further enhances performance;nevertheless,it is challenging due to the nature of small-scale and imbalanced PD datasets.This paper proposed a convolutional neural network-based deep support vector machine(CNN-DSVM)to automate the feature extraction process using CNN and extend the conventional SVM to a DSVM for better classification performance in small-scale PD datasets.A customized kernel function reduces the impact of biased classification towards the majority class(healthy candidates in our consideration).An improved generative adversarial network(IGAN)was designed to generate additional training data to enhance the model’s performance.For performance evaluation,the proposed algorithm achieves a sensitivity of 97.6%and a specificity of 97.3%.The performance comparison is evaluated from five perspectives,including comparisons with different data generation algorithms,feature extraction techniques,kernel functions,and existing works.Results reveal the effectiveness of the IGAN algorithm,which improves the sensitivity and specificity by 4.05%–4.72%and 4.96%–5.86%,respectively;and the effectiveness of the CNN-DSVM algorithm,which improves the sensitivity by 1.24%–57.4%and specificity by 1.04%–163%and reduces biased detection towards the majority class.The ablation experiments confirm the effectiveness of individual components.Two future research directions have also been suggested.
文摘目的:探索通用转录因子ⅡE肽1(GTF2E1)在结直肠癌(CRC)中的表达水平及其对肿瘤细胞增殖的影响,以评估其作为潜在生物标志物和治疗靶点的价值。方法:通过收集并分析来自公共数据库的CRC样本数据,结合临床样本的免疫组化结果,利用单细胞测序技术和成簇规律间隔短回文重复序列(CRISPR)敲除实验,全面评价GTF2E1在CRC中的表达模式和作用。结果:GTF2E1在CRC组织中呈显著高表达,公共数据集、单细胞分析及内部临床样本检测结果一致。其诊断CRC的综合受试者工作特征曲线下面积为0.87,具有良好的诊断效能;CRISPR敲除实验显示,GTF 2 E 1敲除可显著抑制61种CRC细胞系的增殖。结论:GTF2E1在CRC的发生与增殖中起重要作用,有潜力成为CRC的生物标志物和治疗靶点。
