In recent years,neonicotinoids(NEOs)and organophosphate esters(OPEs)have been widely used as substitutes for traditional pesticides and brominated fame-retardants,respectively.Previous studies have shown that those co...In recent years,neonicotinoids(NEOs)and organophosphate esters(OPEs)have been widely used as substitutes for traditional pesticides and brominated fame-retardants,respectively.Previous studies have shown that those compounds can be frequently detected in environmental and human samples,are able to penetrate the placental barrier,and are toxic to animals.Thus,it is reasonable to speculate that NEOs and OPEs may have potential adverse effects in humans,especially during development.We employed a human embryonic stem cell differentiation-and liver S9 fraction metabolism-based fast screening model to assess the potential embryonic toxicity of those two types of chemicals.We show that four NEO and five OPE prototypes targeted mostly ectoderm specification,as neural ectoderm and neural crest genes were down-regulated,and surface ectoderm and placode markers up-regulated.Human liver S9 fraction's treatment could generally reduce the effects of the chemicals,except in a few specific instances,indicating the liver may detoxify NEOs and OPEs.Our findings suggest that NEOs and OPEs interfere with human early embryonic development.展开更多
To solve the problems associated with low resolution and high computational effort infinite time,this paper proposes a fast forward modeling method for muon energy loss transmission tomography based on a model voxeliza...To solve the problems associated with low resolution and high computational effort infinite time,this paper proposes a fast forward modeling method for muon energy loss transmission tomography based on a model voxelization energy loss projection algorithm.First,the energy loss equation for muon transmission tomography is derived from the Bethe–Bloch formula,and the imaging region is then dissected into several units using the model voxelization method.Thereafter,the three-dimensional(3-D)imaging model is discretized into parallel and equally spaced two-dimensional(2-D)slices using the model layering method to realize a dimensional reduction of the 3-D volume data and accelerate the forward calculation speed.Subsequently,the muon energy loss transmission tomography equation is discretized using the ray energy loss projection method to establish a set of energy loss equations for the muon penetration voxel model.Finally,the muon energy loss values at the outgoing point are obtained by solving the projection coefficient matrix of the ray length-weighted model,achieving a significant reduction in the number of muons and improving the computational efficiency.A comparison of our results with the simulation results based on the Monte Carlo method verifies the accuracy and effectiveness of the algorithm proposed in this paper.The metallic mineral identification tests show that the proposed algorithm can quickly identify high-density metallic minerals.The muon energy loss response can accurately identify the boundary of the anomalies and their spatial distribution characteristics.展开更多
A fast version of the State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics （LASG）/Institute of Atmospheric Physics （IAP） climate system model is briefly documented. Th...A fast version of the State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics （LASG）/Institute of Atmospheric Physics （IAP） climate system model is briefly documented. The fast coupled model employs a low resolution version of the atmospheric component Grid Atmospheric Model of IAP/LASG （GAMIL）, with the other parts of the model, namely an oceanic component LASG/IAP Climate Ocean Model （LICOM）, land component Common Land Model （CLM）, and sea ice component from National Center for Atmospheric Research Community Climate System Model （NCAR CCSM2）, as the same as in the standard version of LASG/IAP Flexible Global Ocean Atmosphere Land System model （FGOALS_g）. The parameterizations of physical and dynamical processes of the atmospheric component in the fast version are identical to the standard version, although some parameter values are different. However, by virtue of reduced horizontal resolution and increased time-step of the most time-consuming atmospheric component, it runs faster by a factor of 3 and can serve as a useful tool for longterm and large-ensemble integrations. A 1000-year control simulation of the present-day climate has been completed without flux adjustments. The final 600 years of this simulation has virtually no trends in global mean sea surface temperatures and is recommended for internal variability studies. Several aspects of the control simulation＇s mean climate and variability are evaluated against the observational or reanalysis data. The strengths and weaknesses of the control simulation are evaluated. The mean atmospheric circulation is well simulated, except in high latitudes. The Asian-Australian monsoonal meridional cell shows realistic features, however, an artificial rainfall center is located to the eastern periphery of the Tibetan Plateau persists throughout the year. The mean bias of SST resembles that of the standard version, appearing as a ＂double ITCZ＂ （Inter-Tropical Convergence Zone） associated with a westward extension of the equatorial eastern Pacific cold tongue. The sea ice extent is acceptable but has a higher concentration. The strength of Atlantic meridional overturning is 27.5 Sv. Evidence from the 600-year simulation suggests a modulation of internal variability on ENSO frequency, since both regular and irregular oscillations of ENSO are found during the different time periods of the long-term simulation.展开更多
For the case that two pursuers intercept an evasive target,the cooperative strategies and state estimation methods taken by pursuers can seriously affect the guidance accuracy for the target,which performs a bang For ...For the case that two pursuers intercept an evasive target,the cooperative strategies and state estimation methods taken by pursuers can seriously affect the guidance accuracy for the target,which performs a bang For the case that two pursuers intercept an evasive target,the cooperative strategies and state estimation methods taken by pursuers can seriously affect the guidance accuracy for the target,which performs a bang-bang evasive maneuver with a random switching time.Combined Fast multiple model adaptive estimation(Fast MMAE)algorithm,the cooperative guidance law takes detection configuration affecting the accuracy of interception into consideration.Introduced the detection error model related to the line-of-sight(LOS)separation angle of two interceptors,an optimal cooperative guidance law solving the optimization problem is designed to modulate the LOS separation angle to reduce the estimation error and improve the interception performance.Due to the uncertainty of the target bang-bang maneuver switching time and the effective fitting of its multi-modal motion,Fast MMAE is introduced to identify its maneuver switching time and estimate the acceleration of the target to track and intercept the target accurately.The designed cooperative optimal guidance law with Fast MMAE has better estimation ability and interception performance than the traditional guidance law and estimation method via Monte Carlo simulation.展开更多
The dynamic multichannel binocular visual image modeling is studied based on Internet of Things (IoT) Perception Layer, using mobile robot self-organizing network. By employing multigroup mobile robots with binocular ...The dynamic multichannel binocular visual image modeling is studied based on Internet of Things (IoT) Perception Layer, using mobile robot self-organizing network. By employing multigroup mobile robots with binocular visual system, the real visual images of the object will be obtained. Then through the mobile self-organizing network, a three-dimensional model is rebuilt by synthesizing the returned images. On this basis, we formalize a novel algorithm for multichannel binocular visual three-dimensional images based on fast three-dimensional modeling. Compared with the method based on single binocular visual system, the new algorithm can improve the Integrity and accuracy of the dynamic three-dimensional object modeling. The simulation results show that the new method can effectively accelerate the modeling speed, improve the similarity and not increase the data size.展开更多
This work was focused on the model-based design method of two-axis four-actuator(TAFA) fast steering mirror system(FSM), in order to improve the design efficiency. The structure and operation principle commonality of ...This work was focused on the model-based design method of two-axis four-actuator(TAFA) fast steering mirror system(FSM), in order to improve the design efficiency. The structure and operation principle commonality of normal TAFA FSM were investigated. Based on the structure and the commonality, the conditions of single-axis idea, high-frequency resonance and coupling were modeled gradually. Combining these models, a holonomic system model was established to reflect and predict the performance of TAFA FSM. A model-based design method was proposed based on the holonomic system model. The design flow and design concept of the method were described. In accordance with the method, a TAFA FSM was designed. Simulations and experiments of the FSM were done, and the results of them were compared. The compared results indicate that the holonomic system model can well reflect and predict the performance of TAFA FSM. The bandwidth of TAFA FSM is more than 250 Hz; adjust time is less than 15 ms;overshoot is less than 8%; position accuracy is better than 10 μrad; the FSM prototype can satisfy the requirements.展开更多
This paper attempts to analyze in detail the remote influence of the Indian Ocean Basin warming on the Northwest Pacific （NWP） during the year of decaying E1 Nifio. Observation data and the Fast Ocean- Atmosphere co...This paper attempts to analyze in detail the remote influence of the Indian Ocean Basin warming on the Northwest Pacific （NWP） during the year of decaying E1 Nifio. Observation data and the Fast Ocean- Atmosphere coupled Model 1.5 were used to investigate the triggering conditions under which the remote influence is formed between the positive sea surface temperature （SST） anomaly in the North Indian Ocean and the Anomalous Northwest Pacific anticyclone （ANWPA）. Our research show that it is only when there is a contributory background wind field over the Indian Ocean, i,e., when the Indian Summer Monsoon （ISM） reaches its peak, that the warmer SST anomaly in the North Indian Ocean incites significant easterly wind anomalies in the lower atmosphere of the Indo-West tropical Pacific. This then produces the remote influence on the ANWPA. Therefore, the SST anomaly in the North Indian Ocean might interfere with the prediction of the East Asia Summer Monsoon in the year of decaying E1 Nifio. Both the sustaining effect of local negative SST anomalies in the NWP, and the remote effect of positive SST anomalies in the North Indian Ocean on the ANWPA, should be considered in further research.展开更多
A model that rapidly predicts the density components of raw coal is described.It is based on a threegrade fast float/sink test.The recent comprehensive monthly floating and sinking data are used for comparison.The pre...A model that rapidly predicts the density components of raw coal is described.It is based on a threegrade fast float/sink test.The recent comprehensive monthly floating and sinking data are used for comparison.The predicted data are used to draw washability curves and to provide a rapid evaluation of the effect from heavy medium induced separation.Thirty-one production shifts worth of fast float/sink data and the corresponding quick ash data are used to verify the model.The results show a small error with an arithmetic average of 0.53 and an absolute average error of 1.50.This indicates that this model has high precision.The theoretical yield from the washability curves is 76.47% for the monthly comprehensive data and 81.31% using the model data.This is for a desired cleaned coal ash of 9%.The relative error between these two is 6.33%,which is small and indicates that the predicted data can be used to rapidly evaluate the separation effect of gravity separation equipment.展开更多
Obtaining accurate bathymetric maps is very valuable for marine environment monitoring,port planning,and so on.Accurately estimating water depth in turbid coastal waters using satellite remote sensing encounters chall...Obtaining accurate bathymetric maps is very valuable for marine environment monitoring,port planning,and so on.Accurately estimating water depth in turbid coastal waters using satellite remote sensing encounters challenges originating from low water transparency,but it is limited by the quantity,quality,and water quality of samples.This study introduces a fast feature cascade learning model(FFCLM)to enhance the accuracy of bathymetric inversion from multispectral satellite images,particularly when limited field samples are available.FFCLM leverages spectral bands and in situ data to derive effective inversion weights through feature concatenation and cascade fitting.Field experiments conducted at Nanshan Port and Rushikonda Beach gathered water depth,satellite,and in situ data.Comparative analysis with conventional machine learning algorithms,including support vector machine,random forest,and gradient boosting trees,indicates that FFCLM achieves lower errors and demonstrates more robust performance across study areas.This is especially more pronounced when using small training samples(n<100).Examination of key parameters and water depth profiles highlights FFCLM’s advantages in generalization and deep-water inversion.This study presents an efficient solution for small-sample bathymetric mapping in turbid coastal waters,utilizing spectral and physical information to overcome sample size limitations and enhancing satellite remote sensing capabilities for shallow water monitoring.展开更多
Transformer models have become a cornerstone of various natural language processing(NLP)tasks.However,the substantial computational overhead during the inference remains a significant challenge,limiting their deployme...Transformer models have become a cornerstone of various natural language processing(NLP)tasks.However,the substantial computational overhead during the inference remains a significant challenge,limiting their deployment in practical applications.In this study,we address this challenge by minimizing the inference overhead in transformer models using the controlling element on artificial intelligence(AI)accelerators.Our work is anchored by four key contributions.First,we conduct a comprehensive analysis of the overhead composition within the transformer inference process,identifying the primary bottlenecks.Second,we leverage the management processing element(MPE)of the Shenwei AI(SWAI)accelerator,implementing a three-tier scheduling framework that significantly reduces the number of host-device launches to approximately 1/10000 of the original PyTorch-GPU setup.Third,we introduce a zero-copy memory management technique using segment-page fusion,which significantly reduces memory access latency and improves overall inference efficiency.Finally,we develop a fast model loading method that eliminates redundant computations during model verification and initialization,reducing the total loading time for large models from 22128.31 ms to 1041.72 ms.Our contributions significantly enhance the optimization of transformer models,enabling more efficient and expedited inference processes on AI accelerators.展开更多
Assembly simulations such as assembly process simulation and assembly tolerance simulation have become an effective means to evaluate and analyze product assembly design and assembly process planning.Being core aspec...Assembly simulations such as assembly process simulation and assembly tolerance simulation have become an effective means to evaluate and analyze product assembly design and assembly process planning.Being core aspect of simulation implementation,building an assembly simulation model is rather time-consuming because of its high complexity.Furthermore,modeling has a significant influence on the popularization and application of simulation technology.In this paper,data needed by assembly process and tolerance simulation are addressed to propose a data-driven approach for assembly simulation modeling.The application process and the architecture of modeling framework for assembly simulation are presented as well.An assembly sequence simulation example is given to illustrate the application of the framework.The framework provides a new idea for the realization of automatic modeling for assembly simulation.展开更多
A head-related transfer function （HRTF） model for fast and real-time synthesizing multiple virtual sound sources is proposed. A head-related impulse response （HRIR, time- domain version of HRTF） is first decompose...A head-related transfer function （HRTF） model for fast and real-time synthesizing multiple virtual sound sources is proposed. A head-related impulse response （HRIR, time- domain version of HRTF） is first decomposed by a two-level wavelet packet and then represented by a model composed of subband filters and reconstruction filters. The coefficients of the subband filters are the zero interpolation of the wavelet coefficients of the HRIR. The coefficients of the reconstruction filters can be calculated from the wavelet function. The model is simplified by applying a threshold method to reduce the wavelet coefficients. The calculated results indicate that for a model with 30 wavelet coefficients, the error of reconstructed HRIR is about 1%. And the result of a psychoacoustic test shows that a model with 35 wavelet coefficients is perceptually indistinguishable from the original HRIR. When multiple virtual sound sources are synthesized simultaneously, the computational cost of the proposed model is much less than the traditional HRTF filters.展开更多
基金supported by the Ministry of Science and Technology of the People’s Republic of China (No.2020YFA0907500)the National Natural Science Foundation of China (Nos.22150710514,22021003,and 22106174)+1 种基金the Strategic Priority Research Program of the Chinese Academy of Sciences (No.XDPB200202)the Postdoc Science Foundation of China (No.2021M693322)。
文摘In recent years,neonicotinoids(NEOs)and organophosphate esters(OPEs)have been widely used as substitutes for traditional pesticides and brominated fame-retardants,respectively.Previous studies have shown that those compounds can be frequently detected in environmental and human samples,are able to penetrate the placental barrier,and are toxic to animals.Thus,it is reasonable to speculate that NEOs and OPEs may have potential adverse effects in humans,especially during development.We employed a human embryonic stem cell differentiation-and liver S9 fraction metabolism-based fast screening model to assess the potential embryonic toxicity of those two types of chemicals.We show that four NEO and five OPE prototypes targeted mostly ectoderm specification,as neural ectoderm and neural crest genes were down-regulated,and surface ectoderm and placode markers up-regulated.Human liver S9 fraction's treatment could generally reduce the effects of the chemicals,except in a few specific instances,indicating the liver may detoxify NEOs and OPEs.Our findings suggest that NEOs and OPEs interfere with human early embryonic development.
基金supported by the National Key Research and Development Project of China(2016YFC0303104)the National Natural Science Foundation of China(41304090)。
文摘To solve the problems associated with low resolution and high computational effort infinite time,this paper proposes a fast forward modeling method for muon energy loss transmission tomography based on a model voxelization energy loss projection algorithm.First,the energy loss equation for muon transmission tomography is derived from the Bethe–Bloch formula,and the imaging region is then dissected into several units using the model voxelization method.Thereafter,the three-dimensional(3-D)imaging model is discretized into parallel and equally spaced two-dimensional(2-D)slices using the model layering method to realize a dimensional reduction of the 3-D volume data and accelerate the forward calculation speed.Subsequently,the muon energy loss transmission tomography equation is discretized using the ray energy loss projection method to establish a set of energy loss equations for the muon penetration voxel model.Finally,the muon energy loss values at the outgoing point are obtained by solving the projection coefficient matrix of the ray length-weighted model,achieving a significant reduction in the number of muons and improving the computational efficiency.A comparison of our results with the simulation results based on the Monte Carlo method verifies the accuracy and effectiveness of the algorithm proposed in this paper.The metallic mineral identification tests show that the proposed algorithm can quickly identify high-density metallic minerals.The muon energy loss response can accurately identify the boundary of the anomalies and their spatial distribution characteristics.
基金Acknowledgements. This work was jointly supported by the Chinese Academy of Sciences through the International Partnership Creative Group entitled ＂The Climate System Model Development and Application Studies＂, the Major State Basic Research Development Program of China （973 Program） under Grant No. 2005CB321703, and the National Natural Science Foundation of China （Grant Nos. 40675050, 40221503, 40625014）. The long-term integration of the coupled model was finished on the Lenovo DeepComp 6800 supercomputer at the Supercomputing Center of the Chinese Academy of Sciences, and the IBM SP690 at the Institute of Atmospheric Physics, Chinese Academy of Sciences. The authors appreciate the contribution of Drs. R. C. Yu, Y. Q. Yu, H. L. Liu, W. P. Zheng, J. Li, X. G Xin, and Mrs. H. Wan, H. M. Li in the model development and validations.
文摘A fast version of the State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics （LASG）/Institute of Atmospheric Physics （IAP） climate system model is briefly documented. The fast coupled model employs a low resolution version of the atmospheric component Grid Atmospheric Model of IAP/LASG （GAMIL）, with the other parts of the model, namely an oceanic component LASG/IAP Climate Ocean Model （LICOM）, land component Common Land Model （CLM）, and sea ice component from National Center for Atmospheric Research Community Climate System Model （NCAR CCSM2）, as the same as in the standard version of LASG/IAP Flexible Global Ocean Atmosphere Land System model （FGOALS_g）. The parameterizations of physical and dynamical processes of the atmospheric component in the fast version are identical to the standard version, although some parameter values are different. However, by virtue of reduced horizontal resolution and increased time-step of the most time-consuming atmospheric component, it runs faster by a factor of 3 and can serve as a useful tool for longterm and large-ensemble integrations. A 1000-year control simulation of the present-day climate has been completed without flux adjustments. The final 600 years of this simulation has virtually no trends in global mean sea surface temperatures and is recommended for internal variability studies. Several aspects of the control simulation＇s mean climate and variability are evaluated against the observational or reanalysis data. The strengths and weaknesses of the control simulation are evaluated. The mean atmospheric circulation is well simulated, except in high latitudes. The Asian-Australian monsoonal meridional cell shows realistic features, however, an artificial rainfall center is located to the eastern periphery of the Tibetan Plateau persists throughout the year. The mean bias of SST resembles that of the standard version, appearing as a ＂double ITCZ＂ （Inter-Tropical Convergence Zone） associated with a westward extension of the equatorial eastern Pacific cold tongue. The sea ice extent is acceptable but has a higher concentration. The strength of Atlantic meridional overturning is 27.5 Sv. Evidence from the 600-year simulation suggests a modulation of internal variability on ENSO frequency, since both regular and irregular oscillations of ENSO are found during the different time periods of the long-term simulation.
基金This work was supported by the National Natural Science Foundation(NNSF)of China under grant no.61673386,62073335the China Postdoctoral Science Foundation(2017M613201,2019T120944).
文摘For the case that two pursuers intercept an evasive target,the cooperative strategies and state estimation methods taken by pursuers can seriously affect the guidance accuracy for the target,which performs a bang For the case that two pursuers intercept an evasive target,the cooperative strategies and state estimation methods taken by pursuers can seriously affect the guidance accuracy for the target,which performs a bang-bang evasive maneuver with a random switching time.Combined Fast multiple model adaptive estimation(Fast MMAE)algorithm,the cooperative guidance law takes detection configuration affecting the accuracy of interception into consideration.Introduced the detection error model related to the line-of-sight(LOS)separation angle of two interceptors,an optimal cooperative guidance law solving the optimization problem is designed to modulate the LOS separation angle to reduce the estimation error and improve the interception performance.Due to the uncertainty of the target bang-bang maneuver switching time and the effective fitting of its multi-modal motion,Fast MMAE is introduced to identify its maneuver switching time and estimate the acceleration of the target to track and intercept the target accurately.The designed cooperative optimal guidance law with Fast MMAE has better estimation ability and interception performance than the traditional guidance law and estimation method via Monte Carlo simulation.
基金supported by HiTech Researchand Development Program of China under Grant No.2007AA10Z235
文摘The dynamic multichannel binocular visual image modeling is studied based on Internet of Things (IoT) Perception Layer, using mobile robot self-organizing network. By employing multigroup mobile robots with binocular visual system, the real visual images of the object will be obtained. Then through the mobile self-organizing network, a three-dimensional model is rebuilt by synthesizing the returned images. On this basis, we formalize a novel algorithm for multichannel binocular visual three-dimensional images based on fast three-dimensional modeling. Compared with the method based on single binocular visual system, the new algorithm can improve the Integrity and accuracy of the dynamic three-dimensional object modeling. The simulation results show that the new method can effectively accelerate the modeling speed, improve the similarity and not increase the data size.
基金Projects(51135009)supported by the National Natural Science Foundation of China
文摘This work was focused on the model-based design method of two-axis four-actuator(TAFA) fast steering mirror system(FSM), in order to improve the design efficiency. The structure and operation principle commonality of normal TAFA FSM were investigated. Based on the structure and the commonality, the conditions of single-axis idea, high-frequency resonance and coupling were modeled gradually. Combining these models, a holonomic system model was established to reflect and predict the performance of TAFA FSM. A model-based design method was proposed based on the holonomic system model. The design flow and design concept of the method were described. In accordance with the method, a TAFA FSM was designed. Simulations and experiments of the FSM were done, and the results of them were compared. The compared results indicate that the holonomic system model can well reflect and predict the performance of TAFA FSM. The bandwidth of TAFA FSM is more than 250 Hz; adjust time is less than 15 ms;overshoot is less than 8%; position accuracy is better than 10 μrad; the FSM prototype can satisfy the requirements.
基金Supported by the National Basic Research Program of China(973 Program)(Nos.2010CB428504,2012CB956002)the National Natural Science Foundation of China(Nos.40906005,41105059,41065005,GYHY201106017,GYHY201306027)the National Key Technology Research and Development Program(No.2009BAC51B01)
文摘This paper attempts to analyze in detail the remote influence of the Indian Ocean Basin warming on the Northwest Pacific （NWP） during the year of decaying E1 Nifio. Observation data and the Fast Ocean- Atmosphere coupled Model 1.5 were used to investigate the triggering conditions under which the remote influence is formed between the positive sea surface temperature （SST） anomaly in the North Indian Ocean and the Anomalous Northwest Pacific anticyclone （ANWPA）. Our research show that it is only when there is a contributory background wind field over the Indian Ocean, i,e., when the Indian Summer Monsoon （ISM） reaches its peak, that the warmer SST anomaly in the North Indian Ocean incites significant easterly wind anomalies in the lower atmosphere of the Indo-West tropical Pacific. This then produces the remote influence on the ANWPA. Therefore, the SST anomaly in the North Indian Ocean might interfere with the prediction of the East Asia Summer Monsoon in the year of decaying E1 Nifio. Both the sustaining effect of local negative SST anomalies in the NWP, and the remote effect of positive SST anomalies in the North Indian Ocean on the ANWPA, should be considered in further research.
基金National Natural Science Foundation of China (No. 51174202)Doctoral Fund of Ministry of Education of China (No. 20100095110013)
文摘A model that rapidly predicts the density components of raw coal is described.It is based on a threegrade fast float/sink test.The recent comprehensive monthly floating and sinking data are used for comparison.The predicted data are used to draw washability curves and to provide a rapid evaluation of the effect from heavy medium induced separation.Thirty-one production shifts worth of fast float/sink data and the corresponding quick ash data are used to verify the model.The results show a small error with an arithmetic average of 0.53 and an absolute average error of 1.50.This indicates that this model has high precision.The theoretical yield from the washability curves is 76.47% for the monthly comprehensive data and 81.31% using the model data.This is for a desired cleaned coal ash of 9%.The relative error between these two is 6.33%,which is small and indicates that the predicted data can be used to rapidly evaluate the separation effect of gravity separation equipment.
基金supported by the 2023 Hainan Province“South China Sea New Star”Science and Technology Innovation Talent Platform Project(NHXXRCXM202316)in part by Hainan Natural Science Foundation of China(nos.424QN253 and 620RC602)+5 种基金by the National Natural Science Foundation of China(no.61966013)in part by the Teaching Reform Research Project,Hainan Normal University,hsjg2023-07in part by the National Natural Science Foundation of China under grant 61991454in part by the National Key Research and Development Program of China under grant 2023Y FC3107605in part by the Oceanic Interdisciplinary Program of Shanghai Jiao Tong University under grant SL2022ZD206in part by the Scientific Research Fund of Second Institute of Oceanography,MNR under grant SL2302.
文摘Obtaining accurate bathymetric maps is very valuable for marine environment monitoring,port planning,and so on.Accurately estimating water depth in turbid coastal waters using satellite remote sensing encounters challenges originating from low water transparency,but it is limited by the quantity,quality,and water quality of samples.This study introduces a fast feature cascade learning model(FFCLM)to enhance the accuracy of bathymetric inversion from multispectral satellite images,particularly when limited field samples are available.FFCLM leverages spectral bands and in situ data to derive effective inversion weights through feature concatenation and cascade fitting.Field experiments conducted at Nanshan Port and Rushikonda Beach gathered water depth,satellite,and in situ data.Comparative analysis with conventional machine learning algorithms,including support vector machine,random forest,and gradient boosting trees,indicates that FFCLM achieves lower errors and demonstrates more robust performance across study areas.This is especially more pronounced when using small training samples(n<100).Examination of key parameters and water depth profiles highlights FFCLM’s advantages in generalization and deep-water inversion.This study presents an efficient solution for small-sample bathymetric mapping in turbid coastal waters,utilizing spectral and physical information to overcome sample size limitations and enhancing satellite remote sensing capabilities for shallow water monitoring.
文摘Transformer models have become a cornerstone of various natural language processing(NLP)tasks.However,the substantial computational overhead during the inference remains a significant challenge,limiting their deployment in practical applications.In this study,we address this challenge by minimizing the inference overhead in transformer models using the controlling element on artificial intelligence(AI)accelerators.Our work is anchored by four key contributions.First,we conduct a comprehensive analysis of the overhead composition within the transformer inference process,identifying the primary bottlenecks.Second,we leverage the management processing element(MPE)of the Shenwei AI(SWAI)accelerator,implementing a three-tier scheduling framework that significantly reduces the number of host-device launches to approximately 1/10000 of the original PyTorch-GPU setup.Third,we introduce a zero-copy memory management technique using segment-page fusion,which significantly reduces memory access latency and improves overall inference efficiency.Finally,we develop a fast model loading method that eliminates redundant computations during model verification and initialization,reducing the total loading time for large models from 22128.31 ms to 1041.72 ms.Our contributions significantly enhance the optimization of transformer models,enabling more efficient and expedited inference processes on AI accelerators.
基金the National Science Foundation of China(Grant No.51575031)the Graduate Student Innovation Fund of Beihang University and the key program of the Engineering Research Center of Complex Product Advanced Manufacturing System,Ministry of Education.
文摘Assembly simulations such as assembly process simulation and assembly tolerance simulation have become an effective means to evaluate and analyze product assembly design and assembly process planning.Being core aspect of simulation implementation,building an assembly simulation model is rather time-consuming because of its high complexity.Furthermore,modeling has a significant influence on the popularization and application of simulation technology.In this paper,data needed by assembly process and tolerance simulation are addressed to propose a data-driven approach for assembly simulation modeling.The application process and the architecture of modeling framework for assembly simulation are presented as well.An assembly sequence simulation example is given to illustrate the application of the framework.The framework provides a new idea for the realization of automatic modeling for assembly simulation.
基金supported by the National Nature Science Fund of China(50938003,10774049)State Key Lab of Subtropical Building Science,South China University of Technology
文摘A head-related transfer function （HRTF） model for fast and real-time synthesizing multiple virtual sound sources is proposed. A head-related impulse response （HRIR, time- domain version of HRTF） is first decomposed by a two-level wavelet packet and then represented by a model composed of subband filters and reconstruction filters. The coefficients of the subband filters are the zero interpolation of the wavelet coefficients of the HRIR. The coefficients of the reconstruction filters can be calculated from the wavelet function. The model is simplified by applying a threshold method to reduce the wavelet coefficients. The calculated results indicate that for a model with 30 wavelet coefficients, the error of reconstructed HRIR is about 1%. And the result of a psychoacoustic test shows that a model with 35 wavelet coefficients is perceptually indistinguishable from the original HRIR. When multiple virtual sound sources are synthesized simultaneously, the computational cost of the proposed model is much less than the traditional HRTF filters.
