针对工业装配任务,尤其是不规则轴孔工件装配中,基于学习的前期样本质量低、训练过程不稳定等问题,提出一种融合引斥力模型(Attraction-Repulsion Model,ARM)引导机制和长短期记忆网络(Long Short Term Memory,LSTM)的柔性演员-评论家(S...针对工业装配任务,尤其是不规则轴孔工件装配中,基于学习的前期样本质量低、训练过程不稳定等问题,提出一种融合引斥力模型(Attraction-Repulsion Model,ARM)引导机制和长短期记忆网络(Long Short Term Memory,LSTM)的柔性演员-评论家(Soft Actor-Critic,SAC)算法。首先,为解决训练初期探索效率低的问题,提出一种基于引斥力模型的策略引导机制,通过目标位置信息引导机械臂运动,加速收敛过程;其次,基于长短期记忆网络对算法的策略网络和价值网络进行改进,有效利用历史信息,增强策略学习能力,提高算法的收敛速度和稳定性。仿真结果表明,所提出的算法在行星减速器中心轴装配任务中取得显著的效果,装配成功率高达99.4%,与普通SAC算法相比,平均最大接触力和力矩分别降低了68.8%和79.2%。在物理环境中装配成功率达95%以上,最大接触力和力矩分别小于10 N和1.5 N·m,验证了算法的有效性。展开更多
将联邦学习应用于无线身体区域网络(wireless body area network,WBAN)可解决隐私数据保护问题,但仍然面临着全局模型准确率下降和能耗高的挑战。提出了面向智慧医疗的联邦学习系统模型,构建了各个WBAN节点参与联邦学习的能耗模型,分析...将联邦学习应用于无线身体区域网络(wireless body area network,WBAN)可解决隐私数据保护问题,但仍然面临着全局模型准确率下降和能耗高的挑战。提出了面向智慧医疗的联邦学习系统模型,构建了各个WBAN节点参与联邦学习的能耗模型,分析了其数据特性和资源特性。为保护本地数据隐私,避免直接获取原始数据信息,引入KL散度(Kullback-Leibler divergence)代表各节点的统计异构程度,通过信道增益、带宽等指标代表各节点的系统异构程度,提出了一种结合SAC(soft actor-critic)的联邦学习动态节点选择和资源分配方法。在每轮联邦学习训练开始前,SAC算法根据WBAN节点上传的数据特性和资源特性,动态选择参与训练的节点、分配本地计算资源和通信资源,解决WBAN节点的统计异构性和系统异构性导致的全局模型准确率下降和能耗高的问题。在CIFAR10、FashionMNIST、PathMNIST数据集上的实验表明,所提方法相比FedAvg、FAVOR、FLASH-RL,全局模型准确率至多提高20%、能耗降低了50%,并加快了全局模型收敛速度、减小了准确率波动,证明了所提方法的有效性。展开更多
An in situ hybridization technique for localization of calmodulin(CaM) mRNA in isolated entire embryo sacs and proembryos in Nicotiana tabacum L.cv.W38 has been developed. This technique can be applied to smal...An in situ hybridization technique for localization of calmodulin(CaM) mRNA in isolated entire embryo sacs and proembryos in Nicotiana tabacum L.cv.W38 has been developed. This technique can be applied to small amounts of materials in which a whole view of CaM mRNA distribution can be obtained. The authors revealed that CaM mRNA expression changes dramatically before and after fertilization. Especially interesting is that a prominent CaM mRNA band appears between the egg apparatus and polar nuclei temporarily during the period of pollination and fertilization. The band disappears just prior to fertilization and expands to a fan_shaped region that occupies the micropylar portion of the embryo sac. After fertilization, CaM mRNA accumulates in the elongated zygotes with higher concentration in their chalazal portion than in the micropylar portion. Such an asymmetrical pattern continues to manifest in the early proembryos. It is supposed that CaM mRNA may be involved in the early events and signaling steps associated with double fertilization and zygote polarization in higher plants.展开更多
文摘针对工业装配任务,尤其是不规则轴孔工件装配中,基于学习的前期样本质量低、训练过程不稳定等问题,提出一种融合引斥力模型(Attraction-Repulsion Model,ARM)引导机制和长短期记忆网络(Long Short Term Memory,LSTM)的柔性演员-评论家(Soft Actor-Critic,SAC)算法。首先,为解决训练初期探索效率低的问题,提出一种基于引斥力模型的策略引导机制,通过目标位置信息引导机械臂运动,加速收敛过程;其次,基于长短期记忆网络对算法的策略网络和价值网络进行改进,有效利用历史信息,增强策略学习能力,提高算法的收敛速度和稳定性。仿真结果表明,所提出的算法在行星减速器中心轴装配任务中取得显著的效果,装配成功率高达99.4%,与普通SAC算法相比,平均最大接触力和力矩分别降低了68.8%和79.2%。在物理环境中装配成功率达95%以上,最大接触力和力矩分别小于10 N和1.5 N·m,验证了算法的有效性。
文摘将联邦学习应用于无线身体区域网络(wireless body area network,WBAN)可解决隐私数据保护问题,但仍然面临着全局模型准确率下降和能耗高的挑战。提出了面向智慧医疗的联邦学习系统模型,构建了各个WBAN节点参与联邦学习的能耗模型,分析了其数据特性和资源特性。为保护本地数据隐私,避免直接获取原始数据信息,引入KL散度(Kullback-Leibler divergence)代表各节点的统计异构程度,通过信道增益、带宽等指标代表各节点的系统异构程度,提出了一种结合SAC(soft actor-critic)的联邦学习动态节点选择和资源分配方法。在每轮联邦学习训练开始前,SAC算法根据WBAN节点上传的数据特性和资源特性,动态选择参与训练的节点、分配本地计算资源和通信资源,解决WBAN节点的统计异构性和系统异构性导致的全局模型准确率下降和能耗高的问题。在CIFAR10、FashionMNIST、PathMNIST数据集上的实验表明,所提方法相比FedAvg、FAVOR、FLASH-RL,全局模型准确率至多提高20%、能耗降低了50%,并加快了全局模型收敛速度、减小了准确率波动,证明了所提方法的有效性。
文摘An in situ hybridization technique for localization of calmodulin(CaM) mRNA in isolated entire embryo sacs and proembryos in Nicotiana tabacum L.cv.W38 has been developed. This technique can be applied to small amounts of materials in which a whole view of CaM mRNA distribution can be obtained. The authors revealed that CaM mRNA expression changes dramatically before and after fertilization. Especially interesting is that a prominent CaM mRNA band appears between the egg apparatus and polar nuclei temporarily during the period of pollination and fertilization. The band disappears just prior to fertilization and expands to a fan_shaped region that occupies the micropylar portion of the embryo sac. After fertilization, CaM mRNA accumulates in the elongated zygotes with higher concentration in their chalazal portion than in the micropylar portion. Such an asymmetrical pattern continues to manifest in the early proembryos. It is supposed that CaM mRNA may be involved in the early events and signaling steps associated with double fertilization and zygote polarization in higher plants.
