We show that the central limit theorem for linear statistics over determinantal point processes with J-Hermitian kernels holds under fairly general conditions.In particular,we establish the Gaussian limit for linear s...We show that the central limit theorem for linear statistics over determinantal point processes with J-Hermitian kernels holds under fairly general conditions.In particular,we establish the Gaussian limit for linear statistics over determinantal point processes on the union of two copies of Rdwhen the correlation kernels are J-Hermitian translation-invariant.展开更多
针对现有的作战需求表征与作战方案评价存在不确定性信息、作战方案设计中经验积累与重用以及作战方案推荐中相关性与多样性权衡问题,提出了一种不确定信息下考虑相关性与多样性的作战方案推荐方法。首先,将粗糙集理论融入最优最劣法(be...针对现有的作战需求表征与作战方案评价存在不确定性信息、作战方案设计中经验积累与重用以及作战方案推荐中相关性与多样性权衡问题,提出了一种不确定信息下考虑相关性与多样性的作战方案推荐方法。首先,将粗糙集理论融入最优最劣法(best-worst method,BWM)中对各作战能力属性进行赋权;其次,给出了不确定信息下作战方案与任务需求之间的相关性、作战方案之间的多样性计算模型;然后,提出了权衡相关性和多样性的行列式点过程(determinantal point process,DPP)模型,在此基础上,给出了贪婪最大后验概率(maximum a posterior,MAP)推断算法和贪婪Trade-off推断算法以获得作战方案最优推荐子集;最后,通过案例分析验证了所提模型和方法的适用性和可行性。展开更多
针对大多数的图像自动标注结果中含有冗余标签、信息量不够丰富的问题,本文提出了一种基于级联网络和语义层次结构的图像自动标注方法(CNSH)。首先,输入数据集的图片和标签列表,采用级联的VGG网络提取图像特征,训练条件行列式点过程(DPP...针对大多数的图像自动标注结果中含有冗余标签、信息量不够丰富的问题,本文提出了一种基于级联网络和语义层次结构的图像自动标注方法(CNSH)。首先,输入数据集的图片和标签列表,采用级联的VGG网络提取图像特征,训练条件行列式点过程(DPP)算法模型,计算标签的质量分数确定候选标签列表;其次,利用WordNet检索数据集标签得到语义层次结构和同义词,进而构建加权语义路径;最后,利用DPP算法在候选标签集中采样,得到最终的标注结果。与传统的图像标注任务相比,本文方法得到的标注结果能准确描述图片内容,且不含冗余标签。许多评估指标已用于图像标注和多标签学习,但是它们只专注于评估代表性,忽略了多样性。为了解决上述问题,本文采用了基于语义层次结构的语义指标来共同评估代表性和多样性。在IAPRTC-12和ESP Game 2个基准数据集上的实验表明,与现有方法相比本文方法能够产生更具代表性和多样性的标签。展开更多
近年来,虽然基于生成对抗网络(generative adversarial networks,GAN)的文本生成图像问题取得了很大的突破,它可以根据文本的语义信息生成相应的图像,但是生成的图像结果通常缺乏具体的纹理细节,并且经常出现模式崩塌、缺乏多样性等问...近年来,虽然基于生成对抗网络(generative adversarial networks,GAN)的文本生成图像问题取得了很大的突破,它可以根据文本的语义信息生成相应的图像,但是生成的图像结果通常缺乏具体的纹理细节,并且经常出现模式崩塌、缺乏多样性等问题。针对以上问题,提出一种针对生成对抗网络的行列式点过程(determinant point process for generative adversarial networks,GAN-DPP)方法来提高模型生成样本的质量,并使用StackGAN++、ControlGAN两种基线模型对GAN-DPP进行实现。在训练过程中,该方法使用行列式点过程核矩阵对真实数据和合成数据的多样性进行建模,并通过引入无监督惩罚损失来鼓励生成器生成与真实数据相似的多样性数据,从而提高生成样本的清晰度及多样性,减轻模型崩塌等问题,并且无需增加额外的训练过程。在CUB和Oxford-102数据集上,通过Inception Score、Fréchet Inception Distance分数、Human Rank这3种指标的定量评估,证明了GAN-DPP对生成图像多样性与质量提升的有效性。同时通过定性的可视化比较,证明使用GAN-DPP的模型生成的图像纹理细节更加丰富,多样性显著提高。展开更多
Wireless edge caching has been proposed to reduce data traffic congestion in backhaul links, and it is being envisioned as one of the key components of next-generation wireless networks. This paper focuses on the infl...Wireless edge caching has been proposed to reduce data traffic congestion in backhaul links, and it is being envisioned as one of the key components of next-generation wireless networks. This paper focuses on the influences of different caching strategies in Device-to-Device(D2D) networks. We model the D2D User Equipments(DUEs) as the Gauss determinantal point process considering the repulsion between DUEs, as well as the caching replacement process as a many-to-many matching game. By analyzing existing caching placement strategies, a new caching strategy is proposed, which represents the preference list of DUEs as the ratio of content popularity to cached probability. There are two distinct features in the proposed caching strategy.(1) It can cache other contents besides high popularity contents.(2) It can improve the cache hit ratio and reduce the latency compared with three caching placement strategies: Least Recently Used(LRU), Equal Probability Random Cache(EPRC), and the Most Popular Content Cache(MPC). Meanwhile, we analyze the effect of caching on the system performance in terms of different content popularity factors and cache capacity. Simulation results show that our proposed caching strategy is superior to the three other comparison strategies and can significantly improve the cache hit ratio and reduce the latency.展开更多
基金supported by National Natural Science Foundation of China (Grant Nos.Y7116335K1,11801547 and 11688101)supported by National Natural Science Foundation of China (Grant Nos.11722102 and 12026250)+1 种基金Shanghai Technology Innovation Project (Grant No.21JC1400800)Laboratory of Mathematics for Nonlinear Science,Ministry of Education of China。
文摘We show that the central limit theorem for linear statistics over determinantal point processes with J-Hermitian kernels holds under fairly general conditions.In particular,we establish the Gaussian limit for linear statistics over determinantal point processes on the union of two copies of Rdwhen the correlation kernels are J-Hermitian translation-invariant.
文摘针对现有的作战需求表征与作战方案评价存在不确定性信息、作战方案设计中经验积累与重用以及作战方案推荐中相关性与多样性权衡问题,提出了一种不确定信息下考虑相关性与多样性的作战方案推荐方法。首先,将粗糙集理论融入最优最劣法(best-worst method,BWM)中对各作战能力属性进行赋权;其次,给出了不确定信息下作战方案与任务需求之间的相关性、作战方案之间的多样性计算模型;然后,提出了权衡相关性和多样性的行列式点过程(determinantal point process,DPP)模型,在此基础上,给出了贪婪最大后验概率(maximum a posterior,MAP)推断算法和贪婪Trade-off推断算法以获得作战方案最优推荐子集;最后,通过案例分析验证了所提模型和方法的适用性和可行性。
文摘针对大多数的图像自动标注结果中含有冗余标签、信息量不够丰富的问题,本文提出了一种基于级联网络和语义层次结构的图像自动标注方法(CNSH)。首先,输入数据集的图片和标签列表,采用级联的VGG网络提取图像特征,训练条件行列式点过程(DPP)算法模型,计算标签的质量分数确定候选标签列表;其次,利用WordNet检索数据集标签得到语义层次结构和同义词,进而构建加权语义路径;最后,利用DPP算法在候选标签集中采样,得到最终的标注结果。与传统的图像标注任务相比,本文方法得到的标注结果能准确描述图片内容,且不含冗余标签。许多评估指标已用于图像标注和多标签学习,但是它们只专注于评估代表性,忽略了多样性。为了解决上述问题,本文采用了基于语义层次结构的语义指标来共同评估代表性和多样性。在IAPRTC-12和ESP Game 2个基准数据集上的实验表明,与现有方法相比本文方法能够产生更具代表性和多样性的标签。
文摘近年来,虽然基于生成对抗网络(generative adversarial networks,GAN)的文本生成图像问题取得了很大的突破,它可以根据文本的语义信息生成相应的图像,但是生成的图像结果通常缺乏具体的纹理细节,并且经常出现模式崩塌、缺乏多样性等问题。针对以上问题,提出一种针对生成对抗网络的行列式点过程(determinant point process for generative adversarial networks,GAN-DPP)方法来提高模型生成样本的质量,并使用StackGAN++、ControlGAN两种基线模型对GAN-DPP进行实现。在训练过程中,该方法使用行列式点过程核矩阵对真实数据和合成数据的多样性进行建模,并通过引入无监督惩罚损失来鼓励生成器生成与真实数据相似的多样性数据,从而提高生成样本的清晰度及多样性,减轻模型崩塌等问题,并且无需增加额外的训练过程。在CUB和Oxford-102数据集上,通过Inception Score、Fréchet Inception Distance分数、Human Rank这3种指标的定量评估,证明了GAN-DPP对生成图像多样性与质量提升的有效性。同时通过定性的可视化比较,证明使用GAN-DPP的模型生成的图像纹理细节更加丰富,多样性显著提高。
基金supported by the Fundamental Research Funds for the Central Universities (Nos.FRF-DF-20-12 and FRF-GF-18-017B)。
文摘Wireless edge caching has been proposed to reduce data traffic congestion in backhaul links, and it is being envisioned as one of the key components of next-generation wireless networks. This paper focuses on the influences of different caching strategies in Device-to-Device(D2D) networks. We model the D2D User Equipments(DUEs) as the Gauss determinantal point process considering the repulsion between DUEs, as well as the caching replacement process as a many-to-many matching game. By analyzing existing caching placement strategies, a new caching strategy is proposed, which represents the preference list of DUEs as the ratio of content popularity to cached probability. There are two distinct features in the proposed caching strategy.(1) It can cache other contents besides high popularity contents.(2) It can improve the cache hit ratio and reduce the latency compared with three caching placement strategies: Least Recently Used(LRU), Equal Probability Random Cache(EPRC), and the Most Popular Content Cache(MPC). Meanwhile, we analyze the effect of caching on the system performance in terms of different content popularity factors and cache capacity. Simulation results show that our proposed caching strategy is superior to the three other comparison strategies and can significantly improve the cache hit ratio and reduce the latency.
