查看更多>>摘要：An autonomous altitude adjustment system for a stratospheric satellite(StratoSat)platform is proposed.This platform consists of a helium balloon,a ballonet,and a two-way blower.The helium balloon generates lift to balance the platform gravity.The two-way blower inflates and deflates the ballonet to regulate the buoyancy.Altitude adjustment is achieved by tracking the differential pres-sure difference(DPD),and a threshold switching strategy is used to achieve blower flow control.The vertical acceleration regulation ability is decided not only by the blower flow rate,but also by the designed margin of pressure difference(MPD).Pressure difference is a slow-varying variable compared with altitude,and it is adopted as the control variable.The response speed of the actuator to disturbance can be delayed,and the overshoot caused by the large inertia of the platform is inhibi-ted.This method can maintain a high tracking accuracy and reduce the complexity of model calcula-tion,thus improving the robustness of controller design.
查看更多>>摘要：With the help of pre-trained language models,the accuracy of the entity linking task has made great strides in recent years.However,most models with excellent performance require fine-tuning on a large amount of training data using large pre-trained language models,which is a hardware thresh-old to accomplish this task.Some researchers have achieved competitive results with less training da-ta through ingenious methods,such as utilizing information provided by the named entity recognition model.This paper presents a novel semantic-enhancement-based entity linking approach,named se-mantically enhanced hardware-friendly entity linking(SHEL),which is designed to be hardware friendly and efficient while maintaining good performance.Specifically,SHEL's semantic enhance-ment approach consists of three aspects:(1)semantic compression of entity descriptions using a text summarization model;(2)maximizing the capture of mention contexts using asymmetric heuristics;(3)calculating a fixed size mention representation through pooling operations.These series of seman-tic enhancement methods effectively improve the model's ability to capture semantic information while taking into account the hardware constraints,and significantly improve the model's conver-gence speed by more than 50%compared with the strong baseline model proposed in this paper.In terms of performance,SHEL is comparable to the previous method,with superior performance on six well-established datasets,even though SHEL is trained using a smaller pre-trained language model as the encoder.
HU Zhentao(胡振涛)HU ChonghaoYANG HaoranSHUAI Weiwei...
23-30页
查看更多>>摘要：The unsupervised multi-modal image translation is an emerging domain of computer vision whose goal is to transform an image from the source domain into many diverse styles in the target do-main.However,the multi-generator mechanism is employed among the advanced approaches availa-ble to model different domain mappings,which results in inefficient training of neural networks and pattern collapse,leading to inefficient generation of image diversity.To address this issue,this paper introduces a multi-modal unsupervised image translation framework that uses a generator to perform multi-modal image translation.Specifically,firstly,the domain code is introduced in this paper to explicitly control the different generation tasks.Secondly,this paper brings in the squeeze-and-exci-tation(SE)mechanism and feature attention(FA)module.Finally,the model integrates multiple optimization objectives to ensure efficient multi-modal translation.This paper performs qualitative and quantitative experiments on multiple non-paired benchmark image translation datasets while demon-strating the benefits of the proposed method over existing technologies.Overall,experimental results have shown that the proposed method is versatile and scalable.
DENG Junyong(邓军勇)WANG JunjieJIANG LinXIE Xiaoyan...
31-42页
查看更多>>摘要：Unstructured and irregular graph data causes strong randomness and poor locality of data acces-ses in graph processing.This paper optimizes the depth-branch-resorting algorithm(DBR),and pro-poses a branch-alternation-resorting algorithm(BAR).In order to make the algorithm run in parallel and improve the efficiency of algorithm operation,the BAR algorithm is mapped onto the reconfigu-rable array processor(APR-16)to achieve vertex reordering,effectively improving the locality of graph data.This paper validates the BAR algorithm on the GraphBIG framework,by utilizing the re-ordered dataset with BAR on breadth-first search(BFS),single source shortest paht(SSSP)and betweenness centrality(BC)algorithms for traversal.The results show that compared with DBR and Corder algorithms,BAR can reduce execution time by up to 33.00%,and 51.00%seperatively.In terms of data movement,the BAR algorithm has a maximum reduction of 39.00%compared with the DBR algorithm and 29.66%compared with Corder algorithm.In terms of computational complexity,the BAR algorithm has a maximum reduction of 32.56%compared with DBR algorithm and 53.05%compared with Corder algorithm.
查看更多>>摘要：Given the unconstrained characteristics of the multi-robot coordinated towing system,the rope can only provide a unidirectional constraint force to the suspended object,which leads to the weak ability of the system to resist external disturbances and makes it difficult to control the trajectory of the suspended object.Based on the kinematics and statics of the multi-robot coordinated towing sys-tem with fixed base,the dynamic model of the system is established by using the Newton-Euler equations and the Udwadia-Kalaba equations.To plan the trajectories with high stability and strong control,trajectory planning is performed by combining the dynamics and stability of the towing sys-tem.Based on the dynamic stability of the motion trajectory of the suspended object,the stability of the suspended object is effectively improved through online real-time planning and offline manual ad-justment.The effectiveness of the proposed method is verified by comparing the motion stability of the suspended object before and after planning.The results provide a foundation for the motion plan-ning and coordinated control of the towing system.
查看更多>>摘要：Quantized training has been proven to be a prominent method to achieve deep neural network training under limited computational resources.It uses low bit-width arithmetics with a proper scal-ing factor to achieve negligible accuracy loss.Cambricon-Q is the ASIC design proposed to efficient-ly support quantized training,and achieves significant performance improvement.However,there are still two caveats in the design.First,Cambricon-Q with different hardware specifications may lead to different numerical errors,resulting in non-reproducible behaviors which may become a ma-jor concern in critical applications.Second,Cambricon-Q cannot leverage data sparsity,where con-siderable cycles could still be squeezed out.To address the caveats,the acceleration core of Cambri-con-Q is redesigned to support fine-grained irregular data processing.The new design not only ena-bles acceleration on sparse data,but also enables performing local dynamic quantization by contigu-ous value ranges(which is hardware independent),instead of contiguous addresses(which is de-pendent on hardware factors).Experimental results show that the accuracy loss of the method still keeps negligible,and the accelerator achieves 1.61×performance improvement over Cambricon-Q,with about 10%energy increase.
查看更多>>摘要：Most studies on device-free localization currently focus on single-person scenarios.This paper proposes a novel method for device-free localization that utilizes ZigBee received signal strength indi-cation(RSSI)and a Transformer network structure.The method aims to address the limited re-search and low accuracy of two-person device-free localization.This paper first describes the con-struction of the sensor network used for collecting ZigBee RSSI.It then examines the format and fea-tures of ZigBee data packages.The algorithm design of this paper is then introduced.The box plot method is used to identify abnormal data points,and a neural network is used to establish the map-ping model between ZigBee RSSI matrix and localization coordinates.This neural network includes a Transformer encoder layer as the encoder and a fully connected network as the decoder.The pro-posed method's classification accuracy was experimentally tested in an online test stage,resulting in an accuracy rate of 98.79%.In conclusion,the proposed two-person localization system is novel and has demonstrated high accuracy.
查看更多>>摘要：In order to reduce the intrinsic interference of the filter bank multicarrier-quadrature amplitude modulation(FBMC-QAM)system,a novel filter optimization scheme based on discrete prolate spheroidal sequences(DPSS)is proposed.Firstly,a prototype filter function based on DPSS is de-signed,since the eigenvalue can be used as an indicator of the energy concentration of DPSS,so a threshold is set,and the sequence with the most concentrated energy is selected under the threshold,that is,the sequence with the eigenvalue higher than the threshold,and the prototype filter function is rewritten as a weighted sum function of multiple eigenvectors.Under the energy constraints of the filter,the relationship between the eigenvectors and the intrinsic interference function is established,and the function problem is transformed into an optimization problem for the weighted coefficients.Through the interior point method,the most suitable weight is found to obtain the minimum intrinsic interference result.Theoretical analysis and simulation results show that compared with the prototype filters such as Type1 and CaseC,the DPSS filter applying the proposed optimization algorithm can effectively suppress the intrinsic interference of the system and obtain a better bit error rate(BER)performance.
查看更多>>摘要：The large-scale development of electric vehicles(EVs)requires numerous charging stations to serve them,and the charging stations should be reasonably laid out and planned according to the charging demand of electric vehicles.Considering the costs of both operators and users,a site selec-tion model for optimal layout planning of charging stations is constructed,and a queuing theory ap-proach is used to determine the charging pile configuration to meet the charging demand in the plan-ning area.To solve the difficulties of particle swarm global optimization search,the improved ran-dom drift particle swarm optimization(IRDPSO)and Voronoi diagram are used to jointly solve for the optimal layout of electric vehicles.The final arithmetic analysis verifies the feasibility and practicali-ty of the model and algorithm,and the results show that the total social cost is minimized when the charging station is 9,the location of the charging station is close to the center of gravity and the lay-out is reasonable.
查看更多>>摘要：With the rapid development of the aviation industry,the development of intelligent manufactur-ing equipment represented by composite robots has been paid close attention by the aviation indus-try.Based on the analysis of the background and main structure function of composite robots,this paper focuses on the analysis of key technologies such as composite robot hardware design,visual sensing and planning system,integrated control of'hands,feet,and eyes',multi-robot collabora-tive operation,and safety.The typical applications of composite robots in aviation intelligent manu-facturing such as automatic drilling and connection of aircraft,aircraft surface spraying and finish-ing,parts handling,aircraft measurement,and inspection are presented.The development trends such as standardization of composite robots,integration of'5G + cloud computing +AI',and fu-sion of intelligent sensors are proposed.
万方数据
维普