In large-scaleWireless Rechargeable SensorNetworks(WRSN),traditional forward routingmechanisms often lead to reduced energy efficiency.To address this issue,this paper proposes a WRSN node energy optimization algorith...In large-scaleWireless Rechargeable SensorNetworks(WRSN),traditional forward routingmechanisms often lead to reduced energy efficiency.To address this issue,this paper proposes a WRSN node energy optimization algorithm based on regional partitioning and inter-layer routing.The algorithm employs a dynamic clustering radius method and the K-means clustering algorithm to dynamically partition the WRSN area.Then,the cluster head nodes in the outermost layer select an appropriate layer from the next relay routing region and designate it as the relay layer for data transmission.Relay nodes are selected layer by layer,starting from the outermost cluster heads.Finally,the inter-layer routing mechanism is integrated with regional partitioning and clustering methods to develop the WRSN energy optimization algorithm.To further optimize the algorithm’s performance,we conduct parameter optimization experiments on the relay routing selection function,cluster head rotation energy threshold,and inter-layer relay structure selection,ensuring the best configurations for energy efficiency and network lifespan.Based on these optimizations,simulation results demonstrate that the proposed algorithm outperforms traditional forward routing,K-CHRA,and K-CLP algorithms in terms of node mortality rate and energy consumption,extending the number of rounds to 50%node death by 11.9%,19.3%,and 8.3%in a 500-node network,respectively.展开更多
To maximize the aggregate throughput achieved in heterogeneous networks, this paper investigates inter-session network coding for the distribution of layered source data. We define inter-layer hierarchical random line...To maximize the aggregate throughput achieved in heterogeneous networks, this paper investigates inter-session network coding for the distribution of layered source data. We define inter-layer hierarchical random linear network codes （IHRLNC）, which not only take the flexibility of intersession network coding for layer mixing but also consider the strict priority inherent in the layered source data. Furthermore, we propose the inter-layer hierarchical multicast （IHM）, which performs IHRLNC in the network such that each sink can recover some source layers according to its individu- al capacity. To determine the optimal type of IHRLNC that should be performed on each edge in IHM, we formulate an optimization problem based on 0-1 integer linear programming, and propose a heuristic approach to approximate the optimal solution in polynomial time. Simulation results show that the proposed IHM can achieve throughput gains over the layered muhicast schemes.展开更多
Monolithic three-dimensional integrated circuits(M3D ICs)have emerged as an innovative solution to overcome the limitations of traditional 2D scaling,offering improved performance,reduced power consumption,and enhance...Monolithic three-dimensional integrated circuits(M3D ICs)have emerged as an innovative solution to overcome the limitations of traditional 2D scaling,offering improved performance,reduced power consumption,and enhanced functionality.Inter-layer vias(ILVs),crucial components of M3D ICs,provide vertical connectivity between layers but are susceptible to manufacturing and operational defects,such as stuck-at faults(SAFs),shorts,and opens,which can compromise system reliability.These challenges necessitate advanced built-in self-test(BIST)methodologies to ensure robust fault detection and localization while minimizing the testing overhead.In this paper,we introduce a novel BIST architecture tailored to efficiently detect ILV defects,particularly in irregularly positioned ILVs,and approximately localize them within clusters,using a walking pattern approach.In the proposed BIST framework,ILVs are grouped according to the probability of fault occurrence,enabling efficient detection of all SAFs and bridging faults(BFs)and most multiple faults within each cluster.This strategy empowers designers to fine-tune fault coverage,localization precision,and test duration to meet specific design requirements.The new BIST method addresses a critical shortcoming of existing solutions by significantly reducing the number of test configurations and overall test time using multiple ILV clusters.The method also enhances efficiency in terms of area and hardware utilization,particularly for larger circuit benchmarks.For instance,in the LU32PEENG benchmark,where ILVs are divided into 64 clusters,the power,area,and hardware overheads are minimized to 0.82%,1.03%,and 1.14%,respectively.展开更多
This paper studies the target controllability of multilayer complex networked systems,in which the nodes are highdimensional linear time invariant(LTI)dynamical systems,and the network topology is directed and weighte...This paper studies the target controllability of multilayer complex networked systems,in which the nodes are highdimensional linear time invariant(LTI)dynamical systems,and the network topology is directed and weighted.The influence of inter-layer couplings on the target controllability of multi-layer networks is discussed.It is found that even if there exists a layer which is not target controllable,the entire multi-layer network can still be target controllable due to the inter-layer couplings.For the multi-layer networks with general structure,a necessary and sufficient condition for target controllability is given by establishing the relationship between uncontrollable subspace and output matrix.By the derived condition,it can be found that the system may be target controllable even if it is not state controllable.On this basis,two corollaries are derived,which clarify the relationship between target controllability,state controllability and output controllability.For the multi-layer networks where the inter-layer couplings are directed chains and directed stars,sufficient conditions for target controllability of networked systems are given,respectively.These conditions are easier to verify than the classic criterion.展开更多
Wire-arc additive manufacture(WAAM)has great potential for manufacturing of Al-Cu components.However,inferior mechanical properties of WAAM deposited material restrict its industrial application.Inter-layer cold rolli...Wire-arc additive manufacture(WAAM)has great potential for manufacturing of Al-Cu components.However,inferior mechanical properties of WAAM deposited material restrict its industrial application.Inter-layer cold rolling and thermo-mechanical heat treatment(T8)with pre-stretching deformation between solution and aging treatment were adopted in this study.Their effects on hardness,mechanical properties and microstructure were analyzed and compared to the conventional heat treatment(T6).The results show that cold rolling increases the hardness and strengths,which further increase with T8 treatment.The ultimate tensile strength(UTS)of 513 MPa and yield stress(YS)of 413 MPa can be obtained in the inter-layer cold-rolled sample with T8 treatment,which is much higher than that in the as-deposited samples.The cold-rolled samples show higher elongation than that of as-deposited ones due to significant elimination of porosity in cold rolling;while both the T6 and T8 treatments decrease the elongation.The cold rolling and pre-stretching deformation both contribute to the formation of dense and dispersive precipitatedθ′phases,which inhibits the dislocation movement and enhances the strengths;as a result,T8 treatment shows better strengthening effect than the T6 treatment.The strengthening mechanism was analyzed and it was mainly related to work hardening and precipitation strengthening.展开更多
In multi-layer satellite-terrestrial network, Contact Graph Routing(CGR) uses the contact information among satellites to compute routes. However, due to the resource constraints in satellites, it is extravagant to co...In multi-layer satellite-terrestrial network, Contact Graph Routing(CGR) uses the contact information among satellites to compute routes. However, due to the resource constraints in satellites, it is extravagant to configure lots of the potential contacts into contact plans. What's more, a huge contact plan makes the computing more complex, which further increases computing time. As a result, how to design an efficient contact plan becomes crucial for multi-layer satellite network, which usually has a large scaled topology. In this paper, we propose a distributed contact plan design scheme for multi-layer satellite network by dividing a large contact plan into several partial parts. Meanwhile, a duration based inter-layer contact selection algorithm is proposed to handle contacts disruption problem. The performance of the proposed design was evaluated on our Identifier/Locator split based satellite-terrestrial network testbed with 79 simulation nodes. Experiments showed that the proposed design is able to reduce the data delivery delay.展开更多
Due to the layer-by-layer manufacturing characteristics,metallurgical process of selective laser melting(SLM)is inherently dif-ferent in the building direction because of varying conditions,thereby resulting inter-lay...Due to the layer-by-layer manufacturing characteristics,metallurgical process of selective laser melting(SLM)is inherently dif-ferent in the building direction because of varying conditions,thereby resulting inter-layer heterogeneity.To mitigate such anisotropy,it is of great significance to understand the effects of processing parameters on the property evolution and thus metallurgy of fabrication process.This research proposes one-factor-at-a-time experiment to investigate the influences of laser power and scanning speed on the surface qual-ity,microstructures and mechanical properties of selective laser melted Ti-6Al-4V parts.Surface quality is assessed by roughness around the printings while mechanical properties are evaluated through microhardness and tensile strengths.Phases in microstructure are quantified by XRD to correlate with mechanical properties.Fracture morphology is analyzed to understand the effect of defects and microstructure on mechanical performance.The optimized parameter corresponding to best surface quality and mechanical properties has been found respect-ively in laser power of 190 W and scanning speed of 800 mm/s.After optimization,surface roughness has decreased by 44.47%for upper surface.Yielding strength,tensile strength and elongation rate have improved by 13.17%,43.34%and 64.51%,respectively,with similar hardness and Young’s modulus.In addition,heterogeneity of mechanical properties has great improvement by a range of 31.63%-92.68%.展开更多
Although the inter-layer coupling in layered materials has attracted considerable interest due to its importance in determining physical properties of two- dimensional systems, studies on the inter-layer coupling in o...Although the inter-layer coupling in layered materials has attracted considerable interest due to its importance in determining physical properties of two- dimensional systems, studies on the inter-layer coupling in one-dimensional systems have so far been limited. Double-wall carbon nanotubes （DWCNTs） are one of the most fundamental and ideal model systems to study the interqayer coupling in one-dimensional systems. In this work, Rayleigh scattering spectroscopy and transmission electron microscopy are used to characterize the electronic transition between inner-and outer-nanotubes of the exactly same individual DWCNT. We find that the interqayer coupling is strong, leading to downshifts in most of the optical transition energies （up to -0.2 eV） compared to isolated CNTs. We also find that the presence of metallic tubes lead to stronger shifts. The inter-layer screening of Coulomb interactions is one of the key factors in explaining the observed results.展开更多
As a unique hydro-geological phenomenon in permafrost regions,the seepage of supra-permafrost groundwater will carry a large amount of heat and cause differential settlement in the embankment.This paper presents the r...As a unique hydro-geological phenomenon in permafrost regions,the seepage of supra-permafrost groundwater will carry a large amount of heat and cause differential settlement in the embankment.This paper presents the results of a field study monitoring the supra-permafrost groundwater levels on both sides of an embankment in permafrost regions.It describes a two-dimensional coupled hydro-thermal model and uses it to analyze the influence of seepage on its temperature field considering climate warming.The results show that seepage exacerbates permafrost thawing and thickens the active layer.The thermal influence on the sunny side of the embankment toe is more significant than that on the shady side,which will cause differential settlement in the embankment.After 50 years of operation,the embankment soil temperature with seepage during freezing is 0.2C warmer than that without seepage,and the thermal influence diminished with the increase in depth.Additionally,seepage influences the thermal regime in vertical and horizontal directions of the embankment.During freezing seasons,the thaw depth increases,and the horizontal thaw range decreases.During thawing seasons,the thaw range grows both vertically and horizontally.展开更多
Filter pruning is an important technique to compress convolutional neural networks(CNNs)to acquire light-weight high-performance model for practical deployment.However,the existing filter pruning methods suffer from s...Filter pruning is an important technique to compress convolutional neural networks(CNNs)to acquire light-weight high-performance model for practical deployment.However,the existing filter pruning methods suffer from sharp performance drops when the pruning ratio is large,probably due to the unrecoverable information loss caused by aggressive pruning.In this paper,we propose a dual attention based pruning approach called DualPrune to push the limit of network pruning at an ultra-high compression ratio.Firstly,it adopts a graph attention network(GAT)to automatically extract filter-level and layer-level features from CNNs based on the roles of their filters in the whole computation graph.Then the extracted comprehensive features are fed to a side-attention network,which generates sparse attention weights for individual filters to guide model pruning.To avoid layer collapse,the side-attention network adopts a side-path design to preserve the information flow going through the CNN model properly,which allows the CNN model to be pruned at a high compression ratio at initialization and trained from scratch afterward.Extensive experiments based on several well-known CNN models and real-world datasets show that the proposed DualPrune method outperforms the state-of-the-art methods with significant performance improvement,particularly for model compression at a high pruning ratio.展开更多
Bit-depth scalability is a new research field in the on-going scalable extension of the H.264/AVC（SVC） video coding standard.The key point is to accurately predict the enhancement layer,whose bit depth is 10 or more...Bit-depth scalability is a new research field in the on-going scalable extension of the H.264/AVC（SVC） video coding standard.The key point is to accurately predict the enhancement layer,whose bit depth is 10 or more,from the 8 bit base layer.An improved inter-layer prediction scheme for bit-depth scalability was developed that ensures compatibility with the standard and improves the encoding efficiency.The scheme uses a ＂perfect＂ 8 bit picture with an adaptive neighbor filter whose coefficients are optimized by minimizing the block distortion between the 8 bit reconstructed picture and the ＂perfect＂ picture to achieve a more precise 10 bit prediction based on the filtered picture.Double arithmetic precision is used to further improve the encoding efficiency.Experimental results show that the scheme outperforms the recent joint video team（JVT） proposal in the joint scalable video model（JSVM）.展开更多
Entity relation extraction(ERE)is an important task in the field of information extraction.With the wide application of pre-training language model(PLM)in natural language processing(NLP),using PLM has become a brand ...Entity relation extraction(ERE)is an important task in the field of information extraction.With the wide application of pre-training language model(PLM)in natural language processing(NLP),using PLM has become a brand new research direction of ERE.In this paper,BERT is used to extracting entityrelations,and a separated pipeline architecture is proposed.ERE was decomposed into entity-relation classification sub-task and entity-pair annotation sub-task.Both sub-tasks conduct the pre-training and fine-tuning independently.Combining dynamic and static masking,newVerb-MLM and Entity-MLM BERT pre-training tasks were put forward to enhance the correlation between BERT pre-training and TargetedNLPdownstream task-ERE.Inter-layer sharing attentionmechanismwas added to the model,sharing the attention parameters according to the similarity of the attention matrix.Contrast experiment on the SemEavl 2010 Task8 dataset demonstrates that the new MLM task and inter-layer sharing attention mechanism improve the performance of BERT on the entity relation extraction effectively.展开更多
基金funded by National Natural Science Foundation of China(No.61741303)Guangxi Natural Science Foundation(No.2017GXNSFAA198161)the Foundation Project of Guangxi Key Laboratory of Spatial Information and Mapping(No.21-238-21-16).
文摘In large-scaleWireless Rechargeable SensorNetworks(WRSN),traditional forward routingmechanisms often lead to reduced energy efficiency.To address this issue,this paper proposes a WRSN node energy optimization algorithm based on regional partitioning and inter-layer routing.The algorithm employs a dynamic clustering radius method and the K-means clustering algorithm to dynamically partition the WRSN area.Then,the cluster head nodes in the outermost layer select an appropriate layer from the next relay routing region and designate it as the relay layer for data transmission.Relay nodes are selected layer by layer,starting from the outermost cluster heads.Finally,the inter-layer routing mechanism is integrated with regional partitioning and clustering methods to develop the WRSN energy optimization algorithm.To further optimize the algorithm’s performance,we conduct parameter optimization experiments on the relay routing selection function,cluster head rotation energy threshold,and inter-layer relay structure selection,ensuring the best configurations for energy efficiency and network lifespan.Based on these optimizations,simulation results demonstrate that the proposed algorithm outperforms traditional forward routing,K-CHRA,and K-CLP algorithms in terms of node mortality rate and energy consumption,extending the number of rounds to 50%node death by 11.9%,19.3%,and 8.3%in a 500-node network,respectively.
基金Supported by the National Natural Science Foundation of China （ No. 60832001 ）.
文摘To maximize the aggregate throughput achieved in heterogeneous networks, this paper investigates inter-session network coding for the distribution of layered source data. We define inter-layer hierarchical random linear network codes （IHRLNC）, which not only take the flexibility of intersession network coding for layer mixing but also consider the strict priority inherent in the layered source data. Furthermore, we propose the inter-layer hierarchical multicast （IHM）, which performs IHRLNC in the network such that each sink can recover some source layers according to its individu- al capacity. To determine the optimal type of IHRLNC that should be performed on each edge in IHM, we formulate an optimization problem based on 0-1 integer linear programming, and propose a heuristic approach to approximate the optimal solution in polynomial time. Simulation results show that the proposed IHM can achieve throughput gains over the layered muhicast schemes.
文摘Monolithic three-dimensional integrated circuits(M3D ICs)have emerged as an innovative solution to overcome the limitations of traditional 2D scaling,offering improved performance,reduced power consumption,and enhanced functionality.Inter-layer vias(ILVs),crucial components of M3D ICs,provide vertical connectivity between layers but are susceptible to manufacturing and operational defects,such as stuck-at faults(SAFs),shorts,and opens,which can compromise system reliability.These challenges necessitate advanced built-in self-test(BIST)methodologies to ensure robust fault detection and localization while minimizing the testing overhead.In this paper,we introduce a novel BIST architecture tailored to efficiently detect ILV defects,particularly in irregularly positioned ILVs,and approximately localize them within clusters,using a walking pattern approach.In the proposed BIST framework,ILVs are grouped according to the probability of fault occurrence,enabling efficient detection of all SAFs and bridging faults(BFs)and most multiple faults within each cluster.This strategy empowers designers to fine-tune fault coverage,localization precision,and test duration to meet specific design requirements.The new BIST method addresses a critical shortcoming of existing solutions by significantly reducing the number of test configurations and overall test time using multiple ILV clusters.The method also enhances efficiency in terms of area and hardware utilization,particularly for larger circuit benchmarks.For instance,in the LU32PEENG benchmark,where ILVs are divided into 64 clusters,the power,area,and hardware overheads are minimized to 0.82%,1.03%,and 1.14%,respectively.
基金supported by the National Natural Science Foundation of China (U1808205)Hebei Natural Science Foundation (F2000501005)。
文摘This paper studies the target controllability of multilayer complex networked systems,in which the nodes are highdimensional linear time invariant(LTI)dynamical systems,and the network topology is directed and weighted.The influence of inter-layer couplings on the target controllability of multi-layer networks is discussed.It is found that even if there exists a layer which is not target controllable,the entire multi-layer network can still be target controllable due to the inter-layer couplings.For the multi-layer networks with general structure,a necessary and sufficient condition for target controllability is given by establishing the relationship between uncontrollable subspace and output matrix.By the derived condition,it can be found that the system may be target controllable even if it is not state controllable.On this basis,two corollaries are derived,which clarify the relationship between target controllability,state controllability and output controllability.For the multi-layer networks where the inter-layer couplings are directed chains and directed stars,sufficient conditions for target controllability of networked systems are given,respectively.These conditions are easier to verify than the classic criterion.
基金Project(ZZYJKT2024-08)supported by the State Key Laboratory of Precision Manufacturing for Extreme Service Performance,ChinaProject(2022JB11GX004)supported by Selection of the best Candidates to Undertake Key Research Projects by Dalian City,ChinaProject(201806835007)supported by China Scholarship Council。
文摘Wire-arc additive manufacture(WAAM)has great potential for manufacturing of Al-Cu components.However,inferior mechanical properties of WAAM deposited material restrict its industrial application.Inter-layer cold rolling and thermo-mechanical heat treatment(T8)with pre-stretching deformation between solution and aging treatment were adopted in this study.Their effects on hardness,mechanical properties and microstructure were analyzed and compared to the conventional heat treatment(T6).The results show that cold rolling increases the hardness and strengths,which further increase with T8 treatment.The ultimate tensile strength(UTS)of 513 MPa and yield stress(YS)of 413 MPa can be obtained in the inter-layer cold-rolled sample with T8 treatment,which is much higher than that in the as-deposited samples.The cold-rolled samples show higher elongation than that of as-deposited ones due to significant elimination of porosity in cold rolling;while both the T6 and T8 treatments decrease the elongation.The cold rolling and pre-stretching deformation both contribute to the formation of dense and dispersive precipitatedθ′phases,which inhibits the dislocation movement and enhances the strengths;as a result,T8 treatment shows better strengthening effect than the T6 treatment.The strengthening mechanism was analyzed and it was mainly related to work hardening and precipitation strengthening.
基金supported by National High Technology of China ("863 program") under Grant No. 2015AA015702NSAF under Grant No. U1530118+1 种基金NSFC under Grant No. 61602030National Basic Research Program of China ("973 program") under Grant No. 2013CB329101
文摘In multi-layer satellite-terrestrial network, Contact Graph Routing(CGR) uses the contact information among satellites to compute routes. However, due to the resource constraints in satellites, it is extravagant to configure lots of the potential contacts into contact plans. What's more, a huge contact plan makes the computing more complex, which further increases computing time. As a result, how to design an efficient contact plan becomes crucial for multi-layer satellite network, which usually has a large scaled topology. In this paper, we propose a distributed contact plan design scheme for multi-layer satellite network by dividing a large contact plan into several partial parts. Meanwhile, a duration based inter-layer contact selection algorithm is proposed to handle contacts disruption problem. The performance of the proposed design was evaluated on our Identifier/Locator split based satellite-terrestrial network testbed with 79 simulation nodes. Experiments showed that the proposed design is able to reduce the data delivery delay.
基金Project was supported by the Natural Science Foundation of Fujian Province(Grant No.2020J01873)Science and Technology Major Project of Fujian Province(Grant No.2020HZ03018)+1 种基金Fujian Provincial Foreign Cooperation Project of Science and Technology(Grant No.2020I1003)Fujian Provincial Special Project for Marine Economy Development(Grant No.2021-517).
文摘Due to the layer-by-layer manufacturing characteristics,metallurgical process of selective laser melting(SLM)is inherently dif-ferent in the building direction because of varying conditions,thereby resulting inter-layer heterogeneity.To mitigate such anisotropy,it is of great significance to understand the effects of processing parameters on the property evolution and thus metallurgy of fabrication process.This research proposes one-factor-at-a-time experiment to investigate the influences of laser power and scanning speed on the surface qual-ity,microstructures and mechanical properties of selective laser melted Ti-6Al-4V parts.Surface quality is assessed by roughness around the printings while mechanical properties are evaluated through microhardness and tensile strengths.Phases in microstructure are quantified by XRD to correlate with mechanical properties.Fracture morphology is analyzed to understand the effect of defects and microstructure on mechanical performance.The optimized parameter corresponding to best surface quality and mechanical properties has been found respect-ively in laser power of 190 W and scanning speed of 800 mm/s.After optimization,surface roughness has decreased by 44.47%for upper surface.Yielding strength,tensile strength and elongation rate have improved by 13.17%,43.34%and 64.51%,respectively,with similar hardness and Young’s modulus.In addition,heterogeneity of mechanical properties has great improvement by a range of 31.63%-92.68%.
文摘Although the inter-layer coupling in layered materials has attracted considerable interest due to its importance in determining physical properties of two- dimensional systems, studies on the inter-layer coupling in one-dimensional systems have so far been limited. Double-wall carbon nanotubes （DWCNTs） are one of the most fundamental and ideal model systems to study the interqayer coupling in one-dimensional systems. In this work, Rayleigh scattering spectroscopy and transmission electron microscopy are used to characterize the electronic transition between inner-and outer-nanotubes of the exactly same individual DWCNT. We find that the interqayer coupling is strong, leading to downshifts in most of the optical transition energies （up to -0.2 eV） compared to isolated CNTs. We also find that the presence of metallic tubes lead to stronger shifts. The inter-layer screening of Coulomb interactions is one of the key factors in explaining the observed results.
基金the National Natural Science Foundation of China(Grant No.42001065)Open Project of State Key Laboratory of Frozen Soils Engineering(Grant No.SKLFSE202106)+1 种基金the Natural Science Foundation of Ningxia(2022AAC03052)the University First-Class Discipline Construction Project of Ningxia,China(Grant No.NXYLXK2021A03).
文摘As a unique hydro-geological phenomenon in permafrost regions,the seepage of supra-permafrost groundwater will carry a large amount of heat and cause differential settlement in the embankment.This paper presents the results of a field study monitoring the supra-permafrost groundwater levels on both sides of an embankment in permafrost regions.It describes a two-dimensional coupled hydro-thermal model and uses it to analyze the influence of seepage on its temperature field considering climate warming.The results show that seepage exacerbates permafrost thawing and thickens the active layer.The thermal influence on the sunny side of the embankment toe is more significant than that on the shady side,which will cause differential settlement in the embankment.After 50 years of operation,the embankment soil temperature with seepage during freezing is 0.2C warmer than that without seepage,and the thermal influence diminished with the increase in depth.Additionally,seepage influences the thermal regime in vertical and horizontal directions of the embankment.During freezing seasons,the thaw depth increases,and the horizontal thaw range decreases.During thawing seasons,the thaw range grows both vertically and horizontally.
基金supported by the Natural Science Foundation of Jiangsu Province of China under Grant No.BK20222003the National Natural Science Foundation of China under Grant Nos.61972196,61832008,and 61832005the Collaborative Innovation Center of Novel Software Technology and Industrialization,and the Sino-German Institutes of Social Computing.
文摘Filter pruning is an important technique to compress convolutional neural networks(CNNs)to acquire light-weight high-performance model for practical deployment.However,the existing filter pruning methods suffer from sharp performance drops when the pruning ratio is large,probably due to the unrecoverable information loss caused by aggressive pruning.In this paper,we propose a dual attention based pruning approach called DualPrune to push the limit of network pruning at an ultra-high compression ratio.Firstly,it adopts a graph attention network(GAT)to automatically extract filter-level and layer-level features from CNNs based on the roles of their filters in the whole computation graph.Then the extracted comprehensive features are fed to a side-attention network,which generates sparse attention weights for individual filters to guide model pruning.To avoid layer collapse,the side-attention network adopts a side-path design to preserve the information flow going through the CNN model properly,which allows the CNN model to be pruned at a high compression ratio at initialization and trained from scratch afterward.Extensive experiments based on several well-known CNN models and real-world datasets show that the proposed DualPrune method outperforms the state-of-the-art methods with significant performance improvement,particularly for model compression at a high pruning ratio.
基金Supported by the National Natural Science Foundation of China(No.60772046)the Scientific and Technological Innovation Nurture Funds Major Projects of MOE,China (No.707005)the Intel Research Funds
文摘Bit-depth scalability is a new research field in the on-going scalable extension of the H.264/AVC（SVC） video coding standard.The key point is to accurately predict the enhancement layer,whose bit depth is 10 or more,from the 8 bit base layer.An improved inter-layer prediction scheme for bit-depth scalability was developed that ensures compatibility with the standard and improves the encoding efficiency.The scheme uses a ＂perfect＂ 8 bit picture with an adaptive neighbor filter whose coefficients are optimized by minimizing the block distortion between the 8 bit reconstructed picture and the ＂perfect＂ picture to achieve a more precise 10 bit prediction based on the filtered picture.Double arithmetic precision is used to further improve the encoding efficiency.Experimental results show that the scheme outperforms the recent joint video team（JVT） proposal in the joint scalable video model（JSVM）.
基金Hainan Province High level talent project of basic and applied basic research plan(Natural Science Field)in 2019(No.2019RC100)Haikou City Key Science and Technology Plan Project(2020–049)Hainan Province Key Research and Development Project(ZDYF2020018).
文摘Entity relation extraction(ERE)is an important task in the field of information extraction.With the wide application of pre-training language model(PLM)in natural language processing(NLP),using PLM has become a brand new research direction of ERE.In this paper,BERT is used to extracting entityrelations,and a separated pipeline architecture is proposed.ERE was decomposed into entity-relation classification sub-task and entity-pair annotation sub-task.Both sub-tasks conduct the pre-training and fine-tuning independently.Combining dynamic and static masking,newVerb-MLM and Entity-MLM BERT pre-training tasks were put forward to enhance the correlation between BERT pre-training and TargetedNLPdownstream task-ERE.Inter-layer sharing attentionmechanismwas added to the model,sharing the attention parameters according to the similarity of the attention matrix.Contrast experiment on the SemEavl 2010 Task8 dataset demonstrates that the new MLM task and inter-layer sharing attention mechanism improve the performance of BERT on the entity relation extraction effectively.
