The dynamic routing mechanism in evolvable networks enables adaptive reconfiguration of topol-ogical structures and transmission pathways based on real-time task requirements and data character-istics.However,the heig...The dynamic routing mechanism in evolvable networks enables adaptive reconfiguration of topol-ogical structures and transmission pathways based on real-time task requirements and data character-istics.However,the heightened architectural complexity and expanded parameter dimensionality in evolvable networks present significant implementation challenges when deployed in resource-con-strained environments.Due to the critical paths ignored,traditional pruning strategies cannot get a desired trade-off between accuracy and efficiency.For this reason,a critical path retention pruning(CPRP)method is proposed.By deeply traversing the computational graph,the dependency rela-tionship among nodes is derived.Then the nodes are grouped and sorted according to their contribu-tion value.The redundant operations are removed as much as possible while ensuring that the criti-cal path is not affected.As a result,computational efficiency is improved while a higher accuracy is maintained.On the CIFAR benchmark,the experimental results demonstrate that CPRP-induced pruning incurs accuracy degradation below 4.00%,while outperforming traditional feature-agnostic grouping methods by an average 8.98%accuracy improvement.Simultaneously,the pruned model attains a 2.41 times inference acceleration while achieving 48.92%parameter compression and 53.40%floating-point operations(FLOPs)reduction.展开更多
Breadth-first search(BFS) is an important kernel for graph traversal and has been used by many graph processing applications. Extensive studies have been devoted in boosting the performance of BFS. As the most effecti...Breadth-first search(BFS) is an important kernel for graph traversal and has been used by many graph processing applications. Extensive studies have been devoted in boosting the performance of BFS. As the most effective solution, GPU-acceleration achieves the state-of-the-art result of 3.3×109 traversed edges per second on a NVIDIA Tesla C2050 GPU. A novel vertex frontier based GPU BFS algorithm is proposed, and its main features are three-fold. Firstly, to obtain a better workload balance for irregular graphs, a virtual-queue task decomposition and mapping strategy is introduced for vertex frontier expanding. Secondly, a global deduplicate detection scheme is proposed to remove reduplicative vertices from vertex frontier effectively. Finally, a GPU-based bottom-up BFS approach is employed to process large frontier. The experimental results demonstrate that the algorithm can achieve 10% improvement over the state-of-the-art method on diverse graphs. Especially, it exhibits 2-3 times speedup on low-diameter and scale-free graphs over the state-of-the-art on a NVIDIA Tesla K20 c GPU, reaching a peak traversal rate of 11.2×109 edges/s.展开更多
Many cognitive studies have indicated that the path simplicity may be as important as its distance travelled.However,the optimality of paths for current navigation system is often judged purely on the distance travell...Many cognitive studies have indicated that the path simplicity may be as important as its distance travelled.However,the optimality of paths for current navigation system is often judged purely on the distance travelled or time cost,and not the path simplicity.To balance these factors,this paper presented an algorithm to compute a path that not only possesses fewest turns but also is as short as possible by utilizing the breadth-first-search strategy.The proposed algorithm started searching from a starting point,and expanded layer by layer through searching zero-level reachable points until the endpoint is found,and then deleted unnecessary points in the reverse direction.The forward searching and backward cleaning strategies were presented to build a hierarchical graph of zero-level reachable points,and form a fewestturn-path graph(G^(*)).After that,a classic Dijkstra shortest path algorithm was executed on the G^(*) to obtain a fewestturn-and-shortest path.Comparing with the shortest path in Baidu map,the algorithm in this work has less than half of the turns but the nearly same length.The proposed fewest-turn-and-shortest path algorithm is proved to be more suitable for human beings according to human cognition research.展开更多
With the increasing complexity of distribution network structures originating from the high penetration of renewable energy and responsive loads,fast and accurate fault location technology for distribution networks is...With the increasing complexity of distribution network structures originating from the high penetration of renewable energy and responsive loads,fast and accurate fault location technology for distribution networks is a prerequisite for rapid isolation of faults and restoration of the power supply.In this paper,a fault location method based on community graph depth-first traversal is proposed for fast location of single-phase ground faults in distribution networks.First,this paper defines the fault graph weight of the vertices in the distribution network graph model,which can be used to reflect the topology of the vertices and fault points as well as the fluctuation of the vertices’currents.Then,the vertices on the graph model are clustered by using an improved parallel louvain method(IPLM).Finally,the community formed by IPLM is used as the smallest unit for depth-first traversal to achieve fast and accurate location of the fault section.The paper develops a distribution network graph model of IEEE 33-bus system on the graph database for testing.And three other methods are selected for comparison with IPLMDF.The test results show that IPLMDF can achieve fast and accurate fault location when half of the nodes in the distribution network are equipped with D-PMUs.When some of the D-PMUs lose time synchronization,it is still possible to locate the fault section,and at the same time,the locating results can be avoided by falling into local optimal solutions.展开更多
Grahne et al. have presented a graph algorithm for evaluating a subset of recursive queries. This method consists of two phases. In the first phase, the method transforms a linear binary-chain program into a set of eq...Grahne et al. have presented a graph algorithm for evaluating a subset of recursive queries. This method consists of two phases. In the first phase, the method transforms a linear binary-chain program into a set of equations over expressions containing predicate symbols. In the second phase, a graph is constructed from the equations and the answers are produced by traversing the relevant paths. A new algorithm is described which requires less time than Grahne’ s. The key idea of the improvement is to reduce the search space that will be traversed when a query is invoked. Further, the evaluation of cyclic data is speeded up by generating most answers directly in terms of the answers already found and the associated 'path information' instead of traversing the corresponding paths as usual. In this way, this algorithm achieves a linear time complexity for both acyclic and most of cyclic data.展开更多
In this paper, an optimal method to handle cyclic and acyclic data relations in the linear recursive queries is proposed. High efficiency is achieved by integrating graph traversal mechanisms into a top-down evaluatio...In this paper, an optimal method to handle cyclic and acyclic data relations in the linear recursive queries is proposed. High efficiency is achieved by integrating graph traversal mechanisms into a top-down evaluation. In such a way the subsumption checks and the identification of cyclic data can be done very efficielltly First, based on the subsumption checks, the search space can be reduced drastically by avoiding any redundant expansion operation. In fact, in the case of non-cyclic data, the proposed algorithm requires only linear time for evaluating a linear recursive query. On the other hand, in the case of cyclic data, by using the technique for isolating strongly connected components a lot of answers can be generated directly in terms of the intermediate results and the relevant path information instead of evaluating them by performing algebraic operations. Since the cost of generating an answer is much less than that of evaluating an answer by algebraic operations, the time consumption for cyclic data can be reduced by an order of magnitude or more.展开更多
基金Supported by the National Key Research and Development Program of China(No.2022ZD0119003)and the National Natural Science Founda-tion of China(No.61834005).
文摘The dynamic routing mechanism in evolvable networks enables adaptive reconfiguration of topol-ogical structures and transmission pathways based on real-time task requirements and data character-istics.However,the heightened architectural complexity and expanded parameter dimensionality in evolvable networks present significant implementation challenges when deployed in resource-con-strained environments.Due to the critical paths ignored,traditional pruning strategies cannot get a desired trade-off between accuracy and efficiency.For this reason,a critical path retention pruning(CPRP)method is proposed.By deeply traversing the computational graph,the dependency rela-tionship among nodes is derived.Then the nodes are grouped and sorted according to their contribu-tion value.The redundant operations are removed as much as possible while ensuring that the criti-cal path is not affected.As a result,computational efficiency is improved while a higher accuracy is maintained.On the CIFAR benchmark,the experimental results demonstrate that CPRP-induced pruning incurs accuracy degradation below 4.00%,while outperforming traditional feature-agnostic grouping methods by an average 8.98%accuracy improvement.Simultaneously,the pruned model attains a 2.41 times inference acceleration while achieving 48.92%parameter compression and 53.40%floating-point operations(FLOPs)reduction.
基金Projects(61272142,61103082,61003075,61170261,61103193)supported by the National Natural Science Foundation of ChinaProject supported by the Program for New Century Excellent Talents in University of ChinaProjects(2012AA01A301,2012AA010901)supported by the National High Technology Research and Development Program of China
文摘Breadth-first search(BFS) is an important kernel for graph traversal and has been used by many graph processing applications. Extensive studies have been devoted in boosting the performance of BFS. As the most effective solution, GPU-acceleration achieves the state-of-the-art result of 3.3×109 traversed edges per second on a NVIDIA Tesla C2050 GPU. A novel vertex frontier based GPU BFS algorithm is proposed, and its main features are three-fold. Firstly, to obtain a better workload balance for irregular graphs, a virtual-queue task decomposition and mapping strategy is introduced for vertex frontier expanding. Secondly, a global deduplicate detection scheme is proposed to remove reduplicative vertices from vertex frontier effectively. Finally, a GPU-based bottom-up BFS approach is employed to process large frontier. The experimental results demonstrate that the algorithm can achieve 10% improvement over the state-of-the-art method on diverse graphs. Especially, it exhibits 2-3 times speedup on low-diameter and scale-free graphs over the state-of-the-art on a NVIDIA Tesla K20 c GPU, reaching a peak traversal rate of 11.2×109 edges/s.
基金This research was supported by the National Natural Science Foundation of China(Nos.41471332 and 41101354)the National High Technology Research and Development Program of China(863 Program)(No.2013AA12A302)+1 种基金the Fundamental Research Funds for the Central Universities(No.ZYGX2011J077)the Fund of China Scholarship Council.
文摘Many cognitive studies have indicated that the path simplicity may be as important as its distance travelled.However,the optimality of paths for current navigation system is often judged purely on the distance travelled or time cost,and not the path simplicity.To balance these factors,this paper presented an algorithm to compute a path that not only possesses fewest turns but also is as short as possible by utilizing the breadth-first-search strategy.The proposed algorithm started searching from a starting point,and expanded layer by layer through searching zero-level reachable points until the endpoint is found,and then deleted unnecessary points in the reverse direction.The forward searching and backward cleaning strategies were presented to build a hierarchical graph of zero-level reachable points,and form a fewestturn-path graph(G^(*)).After that,a classic Dijkstra shortest path algorithm was executed on the G^(*) to obtain a fewestturn-and-shortest path.Comparing with the shortest path in Baidu map,the algorithm in this work has less than half of the turns but the nearly same length.The proposed fewest-turn-and-shortest path algorithm is proved to be more suitable for human beings according to human cognition research.
基金supported by the National Natural Science Foundation of China (Grant Nos.52009106,51779206)the National Key R&D Program of China (No.2018YFB1500800)the Natural Science Fund Youth Project of Shaanxi Province (2019J-130).
文摘With the increasing complexity of distribution network structures originating from the high penetration of renewable energy and responsive loads,fast and accurate fault location technology for distribution networks is a prerequisite for rapid isolation of faults and restoration of the power supply.In this paper,a fault location method based on community graph depth-first traversal is proposed for fast location of single-phase ground faults in distribution networks.First,this paper defines the fault graph weight of the vertices in the distribution network graph model,which can be used to reflect the topology of the vertices and fault points as well as the fluctuation of the vertices’currents.Then,the vertices on the graph model are clustered by using an improved parallel louvain method(IPLM).Finally,the community formed by IPLM is used as the smallest unit for depth-first traversal to achieve fast and accurate location of the fault section.The paper develops a distribution network graph model of IEEE 33-bus system on the graph database for testing.And three other methods are selected for comparison with IPLMDF.The test results show that IPLMDF can achieve fast and accurate fault location when half of the nodes in the distribution network are equipped with D-PMUs.When some of the D-PMUs lose time synchronization,it is still possible to locate the fault section,and at the same time,the locating results can be avoided by falling into local optimal solutions.
文摘Grahne et al. have presented a graph algorithm for evaluating a subset of recursive queries. This method consists of two phases. In the first phase, the method transforms a linear binary-chain program into a set of equations over expressions containing predicate symbols. In the second phase, a graph is constructed from the equations and the answers are produced by traversing the relevant paths. A new algorithm is described which requires less time than Grahne’ s. The key idea of the improvement is to reduce the search space that will be traversed when a query is invoked. Further, the evaluation of cyclic data is speeded up by generating most answers directly in terms of the answers already found and the associated 'path information' instead of traversing the corresponding paths as usual. In this way, this algorithm achieves a linear time complexity for both acyclic and most of cyclic data.
文摘In this paper, an optimal method to handle cyclic and acyclic data relations in the linear recursive queries is proposed. High efficiency is achieved by integrating graph traversal mechanisms into a top-down evaluation. In such a way the subsumption checks and the identification of cyclic data can be done very efficielltly First, based on the subsumption checks, the search space can be reduced drastically by avoiding any redundant expansion operation. In fact, in the case of non-cyclic data, the proposed algorithm requires only linear time for evaluating a linear recursive query. On the other hand, in the case of cyclic data, by using the technique for isolating strongly connected components a lot of answers can be generated directly in terms of the intermediate results and the relevant path information instead of evaluating them by performing algebraic operations. Since the cost of generating an answer is much less than that of evaluating an answer by algebraic operations, the time consumption for cyclic data can be reduced by an order of magnitude or more.
