针对嵌入式平台下卷积神经网络运行速度慢,无法快速手势检测的问题,提出一种基于SSD的卷积神经网络的嵌入式手势检测算法,该算法显著提高了手势检测速度,并保持了高精度。首先通过一种预处理方法,对原来的手势数据库进行5倍扩展;然后对...针对嵌入式平台下卷积神经网络运行速度慢,无法快速手势检测的问题,提出一种基于SSD的卷积神经网络的嵌入式手势检测算法,该算法显著提高了手势检测速度,并保持了高精度。首先通过一种预处理方法,对原来的手势数据库进行5倍扩展;然后对SSD算法的基础神经网络层进行卷积因子分解,使用MobileNet神经网络获得了在CPU下的3倍加速;最后通过改变输入图片大小同时改变网络结构,减少了算法的计算复杂度。实验结果表明所提算法在两个数据集上的平均精度均值(Mean Average Precision,mAP)下降2.7%,但是在Qualcomm SnapDragon820平台下检测一张图片时间可达到0.233 s,检测速度提高40倍以上。展开更多
卷积神经网络作为深度学习的重要分支,在图像识别、图像分类等方面有广泛的应用,其中快速特征嵌入卷积神经网络框架(convolutional architecture for fast feature embedding,Caffe)是目前炙手可热的深度学习工具.针对固定群体中的目标...卷积神经网络作为深度学习的重要分支,在图像识别、图像分类等方面有广泛的应用,其中快速特征嵌入卷积神经网络框架(convolutional architecture for fast feature embedding,Caffe)是目前炙手可热的深度学习工具.针对固定群体中的目标人物,提出一种基于卷积神经网络的分类方法,该方法不依赖于人脸图像集,而是通过摄像头采集视频,并利用直方图的归一化互相关方法从视频中截取训练图片,再通过Caffe产生训练模型,并将个体目标图片在模型中进行匹配,达到在固定人物群体中对个体目标进行分类的目的.实验结果表明,利用前期的训练模型可对固定群体中的个体目标进行准确匹配.展开更多
Machine learning potentials are promising in atomistic simulations due to their comparable accuracy to first-principles theory but much lower computational cost.However,the reliability,speed,and transferability of ato...Machine learning potentials are promising in atomistic simulations due to their comparable accuracy to first-principles theory but much lower computational cost.However,the reliability,speed,and transferability of atomistic machine learning potentials depend strongly on the way atomic configurations are represented.A wise choice of descriptors used as input for the machine learning program is the key for a successful machine learning representation.Here we develop a simple and efficient strategy to automatically select an optimal set of linearly-independent atomic features out of a large pool of candidates,based on the correlations that are intrinsic to the training data.Through applications to the construction of embedded atom neural network potentials for several benchmark molecules with less redundant linearly-independent embedded density descriptors,we demonstrate the efficiency and accuracy of this new strategy.The proposed algorithm can greatly simplify the initial selection of atomic features and vastly improve the performance of the atomistic machine learning potentials.展开更多
To solve the low efficiency of approximate queries caused by the large sizes of the knowledge graphs in the real world,an embedding-based approximate query method is proposed.First,the nodes in the query graph are cla...To solve the low efficiency of approximate queries caused by the large sizes of the knowledge graphs in the real world,an embedding-based approximate query method is proposed.First,the nodes in the query graph are classified according to the degrees of approximation required for different types of nodes.This classification transforms the query problem into three constraints,from which approximate information is extracted.Second,candidates are generated by calculating the similarity between embeddings.Finally,a deep neural network model is designed,incorporating a loss function based on the high-dimensional ellipsoidal diffusion distance.This model identifies the distance between nodes using their embeddings and constructs a score function.k nodes are returned as the query results.The results show that the proposed method can return both exact results and approximate matching results.On datasets DBLP(DataBase systems and Logic Programming)and FUA-S(Flight USA Airports-Sparse),this method exhibits superior performance in terms of precision and recall,returning results in 0.10 and 0.03 s,respectively.This indicates greater efficiency compared to PathSim and other comparative methods.展开更多
文摘针对嵌入式平台下卷积神经网络运行速度慢,无法快速手势检测的问题,提出一种基于SSD的卷积神经网络的嵌入式手势检测算法,该算法显著提高了手势检测速度,并保持了高精度。首先通过一种预处理方法,对原来的手势数据库进行5倍扩展;然后对SSD算法的基础神经网络层进行卷积因子分解,使用MobileNet神经网络获得了在CPU下的3倍加速;最后通过改变输入图片大小同时改变网络结构,减少了算法的计算复杂度。实验结果表明所提算法在两个数据集上的平均精度均值(Mean Average Precision,mAP)下降2.7%,但是在Qualcomm SnapDragon820平台下检测一张图片时间可达到0.233 s,检测速度提高40倍以上。
文摘卷积神经网络作为深度学习的重要分支,在图像识别、图像分类等方面有广泛的应用,其中快速特征嵌入卷积神经网络框架(convolutional architecture for fast feature embedding,Caffe)是目前炙手可热的深度学习工具.针对固定群体中的目标人物,提出一种基于卷积神经网络的分类方法,该方法不依赖于人脸图像集,而是通过摄像头采集视频,并利用直方图的归一化互相关方法从视频中截取训练图片,再通过Caffe产生训练模型,并将个体目标图片在模型中进行匹配,达到在固定人物群体中对个体目标进行分类的目的.实验结果表明,利用前期的训练模型可对固定群体中的个体目标进行准确匹配.
基金supported by CAS Project for Young Scientists in Basic Research(YSBR-005)the National Natural Science Foundation of China(No.22073089 and No.22033007)+1 种基金Anhui Initiative in Quantum Information Technologies(AHY090200)the Fundamental Research Funds for Central Universities(WK2060000017)。
文摘Machine learning potentials are promising in atomistic simulations due to their comparable accuracy to first-principles theory but much lower computational cost.However,the reliability,speed,and transferability of atomistic machine learning potentials depend strongly on the way atomic configurations are represented.A wise choice of descriptors used as input for the machine learning program is the key for a successful machine learning representation.Here we develop a simple and efficient strategy to automatically select an optimal set of linearly-independent atomic features out of a large pool of candidates,based on the correlations that are intrinsic to the training data.Through applications to the construction of embedded atom neural network potentials for several benchmark molecules with less redundant linearly-independent embedded density descriptors,we demonstrate the efficiency and accuracy of this new strategy.The proposed algorithm can greatly simplify the initial selection of atomic features and vastly improve the performance of the atomistic machine learning potentials.
基金The State Grid Technology Project(No.5108202340042A-1-1-ZN).
文摘To solve the low efficiency of approximate queries caused by the large sizes of the knowledge graphs in the real world,an embedding-based approximate query method is proposed.First,the nodes in the query graph are classified according to the degrees of approximation required for different types of nodes.This classification transforms the query problem into three constraints,from which approximate information is extracted.Second,candidates are generated by calculating the similarity between embeddings.Finally,a deep neural network model is designed,incorporating a loss function based on the high-dimensional ellipsoidal diffusion distance.This model identifies the distance between nodes using their embeddings and constructs a score function.k nodes are returned as the query results.The results show that the proposed method can return both exact results and approximate matching results.On datasets DBLP(DataBase systems and Logic Programming)and FUA-S(Flight USA Airports-Sparse),this method exhibits superior performance in terms of precision and recall,returning results in 0.10 and 0.03 s,respectively.This indicates greater efficiency compared to PathSim and other comparative methods.
