针对深度神经网络(deep neural network,DNN)模型在传统切片与映射方法中存在的资源调度和数据传输瓶颈问题,提出了一种基于片上网络(network on chip,NoC)加速器的高效DNN动态切片与智能映射优化算法。该算法通过动态切片技术灵活划分...针对深度神经网络(deep neural network,DNN)模型在传统切片与映射方法中存在的资源调度和数据传输瓶颈问题,提出了一种基于片上网络(network on chip,NoC)加速器的高效DNN动态切片与智能映射优化算法。该算法通过动态切片技术灵活划分DNN模型的计算任务,并结合智能映射策略优化NoC架构中的任务分配与数据流管理。实验结果表明,与传统方法相比,该算法在计算吞吐量、NoC传输时延、外部内存访问次数和计算能效等方面均显著提升,尤其在复杂模型上表现突出。展开更多
针对不同应用场景的用户利用底层网络资源不充分的问题,提出一种利用网络切片技术对切片进行准入控制和资源分配联合算法(Joint Access Control and Resource Allocation Algorithm for Slicing,JACRAAS)。在第五代移动通信技术(5th Gen...针对不同应用场景的用户利用底层网络资源不充分的问题,提出一种利用网络切片技术对切片进行准入控制和资源分配联合算法(Joint Access Control and Resource Allocation Algorithm for Slicing,JACRAAS)。在第五代移动通信技术(5th Generation Mobile Communication Technology,5G)的演进(5G-Advanced,5G-A)标准下,通过最大化网络切片提供商(Network Slicing Provider,NSP)的收益,使用双深度Q网络算法对网络切片请求进行智能高效的准入控制和资源分配,并对重要经验优先回放,拒绝不满足条件的切片请求。同时,考虑网络拓扑对节点的影响,对重要节点优先排序,并进行节点映射和链路映射。仿真结果表明,所提算法与深度Q网络算法和Q学习算法相比,NSP收益成本比分别提高了9%和15%,资源利用率分别提升了10%和14%,所提算法可以显著提高底层资源的利用率。展开更多
The photodissociation dynamics of isocyanic acid （HNCO） has been studied by the time- sliced velocity map ion imaging technique at 193 nm. The NH（a1△） products were measured via （2＋1） resonance enhanced multip...The photodissociation dynamics of isocyanic acid （HNCO） has been studied by the time- sliced velocity map ion imaging technique at 193 nm. The NH（a1△） products were measured via （2＋1） resonance enhanced multiphoton ionization. Images have been accumulated for the NH（a1△） rotational states in the ground and vibrational excited state （v=0 and 1）. The center-of-mass translational energy distribution derived from the NH（a1△） images implies that the CO vibrational distributions are inverted for most of the measured 1NH（v｜j） internal states. The anisotropic product angular distribution observed indicates a rapid dissociation process for the N-C bond cleavage. A bimodal rotational state distribution of CO（v） has been observed, this result implies that isocyanic acid dissociates in the S1 state in two different pathways.展开更多
文摘针对深度神经网络(deep neural network,DNN)模型在传统切片与映射方法中存在的资源调度和数据传输瓶颈问题,提出了一种基于片上网络(network on chip,NoC)加速器的高效DNN动态切片与智能映射优化算法。该算法通过动态切片技术灵活划分DNN模型的计算任务,并结合智能映射策略优化NoC架构中的任务分配与数据流管理。实验结果表明,与传统方法相比,该算法在计算吞吐量、NoC传输时延、外部内存访问次数和计算能效等方面均显著提升,尤其在复杂模型上表现突出。
文摘针对不同应用场景的用户利用底层网络资源不充分的问题,提出一种利用网络切片技术对切片进行准入控制和资源分配联合算法(Joint Access Control and Resource Allocation Algorithm for Slicing,JACRAAS)。在第五代移动通信技术(5th Generation Mobile Communication Technology,5G)的演进(5G-Advanced,5G-A)标准下,通过最大化网络切片提供商(Network Slicing Provider,NSP)的收益,使用双深度Q网络算法对网络切片请求进行智能高效的准入控制和资源分配,并对重要经验优先回放,拒绝不满足条件的切片请求。同时,考虑网络拓扑对节点的影响,对重要节点优先排序,并进行节点映射和链路映射。仿真结果表明,所提算法与深度Q网络算法和Q学习算法相比,NSP收益成本比分别提高了9%和15%,资源利用率分别提升了10%和14%,所提算法可以显著提高底层资源的利用率。
基金supported by the National Natural Science Foundation of China(No.21573227,N0.11604052)the National Science Foundation of Anhui Province of China(No.1608085QA19)+2 种基金the Natural Science Research Project of Education Department of Anhui Province of China(No.2014KJ020)the Open Foundation of State Key Laboratory(No.SKLMRDK201503,No.SKLMRD-K201611,and No.SKLMRDK201711)the Doctoral Foundation of Fuyang Normal University(No.FSB201501005)
文摘The photodissociation dynamics of isocyanic acid （HNCO） has been studied by the time- sliced velocity map ion imaging technique at 193 nm. The NH（a1△） products were measured via （2＋1） resonance enhanced multiphoton ionization. Images have been accumulated for the NH（a1△） rotational states in the ground and vibrational excited state （v=0 and 1）. The center-of-mass translational energy distribution derived from the NH（a1△） images implies that the CO vibrational distributions are inverted for most of the measured 1NH（v｜j） internal states. The anisotropic product angular distribution observed indicates a rapid dissociation process for the N-C bond cleavage. A bimodal rotational state distribution of CO（v） has been observed, this result implies that isocyanic acid dissociates in the S1 state in two different pathways.
