Using the blocking techniques and m-dependent methods,the asymptotic behavior of kernel density estimators for a class of stationary processes,which includes some nonlinear time series models,is investigated.First,the...Using the blocking techniques and m-dependent methods,the asymptotic behavior of kernel density estimators for a class of stationary processes,which includes some nonlinear time series models,is investigated.First,the pointwise and uniformly weak convergence rates of the deviation of kernel density estimator with respect to its mean(and the true density function)are derived.Secondly,the corresponding strong convergence rates are investigated.It is showed,under mild conditions on the kernel functions and bandwidths,that the optimal rates for the i.i.d.density models are also optimal for these processes.展开更多
It is demonstrated that “survival of the fittest” approach suffers fundamental flaw planted in its very goal: reaching a uniform state starting from a minor random event. Simple considerations prove that a generic p...It is demonstrated that “survival of the fittest” approach suffers fundamental flaw planted in its very goal: reaching a uniform state starting from a minor random event. Simple considerations prove that a generic property of any such state is its global instability. That is why a new approach to the evolution is put forward. It conjectures equilibrium for systems put in an ever-changing environment. The importance of this issue lies in the view that an ever-changing environment is much closer to the natural environment where the biological species live in. The major goal of the present paper is to demonstrate that a specific form of dynamical equilibrium among certain mutations is established in each and every stable in a long-run system. Major result of our considerations is that neither mutation nor either kind dominates forever because a temporary dynamical equilibrium is replaced with another one in the time course. It will be demonstrated that the evolution of those pieces of equilibrium is causal, yet not predetermined process.展开更多
Relativistic diffraction in time wave functions can be used as a basis for causal scattering waves. We derive such exact wave function for a beam of Dirac and Klein-Gordon particles. The transient Dirac spinors are ex...Relativistic diffraction in time wave functions can be used as a basis for causal scattering waves. We derive such exact wave function for a beam of Dirac and Klein-Gordon particles. The transient Dirac spinors are expressed in terms of integral defined functions which are the relativistic equivalent of the Fresnel integrals. When plotted versus time the exact relativistic densities show transient oscillations which resemble a diffraction pattern. The Dirac and Klein-Gordon time oscillations look different, hence relativistic diffraction in time depends strongly on the particle spin.展开更多
领域自适应旨在利用带标签的源域数据和无标签的目标域数据来解决机器学习泛化性不足的问题.现有领域自适应工作主要针对计算机视觉任务.为了解决针对时间序列数据的领域自适应挑战,现有的方法将针对图片数据的方法直接应用于时间序列...领域自适应旨在利用带标签的源域数据和无标签的目标域数据来解决机器学习泛化性不足的问题.现有领域自适应工作主要针对计算机视觉任务.为了解决针对时间序列数据的领域自适应挑战,现有的方法将针对图片数据的方法直接应用于时间序列数据中.这些方法虽然一定程度上解决了模型的泛化能力,但是这些方法依然不能很好地提取解耦的领域不变的特征,从而使得模型的泛化性能依然不尽人意.为了解决这个挑战,提出基于隐变量解耦学习的无监督领域自适应算法.首先,提出针对时间序列数据的因果数据生成过程,在这个生成过程中,假设观测数据背后的隐变量分为变化部分和不变部分,并且将这些部分用隐变量表示.基于这个数据生成过程,提出可识别性理论证明领域变化的隐变量是可以被识别的.在可识别性理论的基础上,设计针对时间序列的隐变量解耦学习领域自适应模型(time series domain adaptation via disentangling invariant and variant latent variables,DIVV).该模型一方面利用变分推断解耦领域变化的隐变量,另一方面采用基于正交特征的对齐模块以解耦领域不变的隐变量.最后该模型采用领域不变特征进行时间序列分类.在多个真实数据集上进行验证,并且取得了最有效的实验结果,证明所提理论和模型在真实场景中的有效性.展开更多
基金supported by National Natural Science Foundation of China(Grant Nos.11171303 and 61273093)the Specialized Research Fund for the Doctor Program of Higher Education(Grant No.20090101110020)
文摘Using the blocking techniques and m-dependent methods,the asymptotic behavior of kernel density estimators for a class of stationary processes,which includes some nonlinear time series models,is investigated.First,the pointwise and uniformly weak convergence rates of the deviation of kernel density estimator with respect to its mean(and the true density function)are derived.Secondly,the corresponding strong convergence rates are investigated.It is showed,under mild conditions on the kernel functions and bandwidths,that the optimal rates for the i.i.d.density models are also optimal for these processes.
文摘It is demonstrated that “survival of the fittest” approach suffers fundamental flaw planted in its very goal: reaching a uniform state starting from a minor random event. Simple considerations prove that a generic property of any such state is its global instability. That is why a new approach to the evolution is put forward. It conjectures equilibrium for systems put in an ever-changing environment. The importance of this issue lies in the view that an ever-changing environment is much closer to the natural environment where the biological species live in. The major goal of the present paper is to demonstrate that a specific form of dynamical equilibrium among certain mutations is established in each and every stable in a long-run system. Major result of our considerations is that neither mutation nor either kind dominates forever because a temporary dynamical equilibrium is replaced with another one in the time course. It will be demonstrated that the evolution of those pieces of equilibrium is causal, yet not predetermined process.
文摘Relativistic diffraction in time wave functions can be used as a basis for causal scattering waves. We derive such exact wave function for a beam of Dirac and Klein-Gordon particles. The transient Dirac spinors are expressed in terms of integral defined functions which are the relativistic equivalent of the Fresnel integrals. When plotted versus time the exact relativistic densities show transient oscillations which resemble a diffraction pattern. The Dirac and Klein-Gordon time oscillations look different, hence relativistic diffraction in time depends strongly on the particle spin.
文摘领域自适应旨在利用带标签的源域数据和无标签的目标域数据来解决机器学习泛化性不足的问题.现有领域自适应工作主要针对计算机视觉任务.为了解决针对时间序列数据的领域自适应挑战,现有的方法将针对图片数据的方法直接应用于时间序列数据中.这些方法虽然一定程度上解决了模型的泛化能力,但是这些方法依然不能很好地提取解耦的领域不变的特征,从而使得模型的泛化性能依然不尽人意.为了解决这个挑战,提出基于隐变量解耦学习的无监督领域自适应算法.首先,提出针对时间序列数据的因果数据生成过程,在这个生成过程中,假设观测数据背后的隐变量分为变化部分和不变部分,并且将这些部分用隐变量表示.基于这个数据生成过程,提出可识别性理论证明领域变化的隐变量是可以被识别的.在可识别性理论的基础上,设计针对时间序列的隐变量解耦学习领域自适应模型(time series domain adaptation via disentangling invariant and variant latent variables,DIVV).该模型一方面利用变分推断解耦领域变化的隐变量,另一方面采用基于正交特征的对齐模块以解耦领域不变的隐变量.最后该模型采用领域不变特征进行时间序列分类.在多个真实数据集上进行验证,并且取得了最有效的实验结果,证明所提理论和模型在真实场景中的有效性.
