时变动态贝叶斯网络模型及其在皮层脑电网络连接中的应用

大脑执行语言的发音需要顶叶、颞叶、额叶等多个脑区协同完成.皮层脑电具有高时间分辨率、较高空间分辨率和高信噪比等优势,为研究大脑的电生理特性提供了重要的技术手段.为了探索大脑对语言的动态处理过程,利用多尺度皮层脑电(标准电极与微电极)分析了被试在执行音节朗读任务时的皮层脑电信号的高频gamma段特征,提出采用时变动态贝叶斯网络构建单次实验任务的有向网络.结果显示该方法能够快速有效地构建语言任务过程中标准电极、微电极以及二者之间的有向网络连接,且反映了大规模网络(标准电极之间的连接)、局部网络(微电极之间的连接)以及大规模网络与局部网络之间的连接(标准电极与微电极之间的连接)随语言任务发生的动态改变.研究还发现,发音时刻之前与之后的网络连接存在显著性差异,且发音方式不同的音节网络间也存在明显差异.该研究将有助于癫痫等神经疾病的术前临床评估以及理解大脑对语言加工的实时处理过程.     

Critical theories of phase transition between symmetry protected topological states and their relation to the gapless boundary theories

Symmetry protected topological states (SPTs) have the same symmetry and the phase transition between them are beyond Landau?s symmetry breaking formalism. In this paper we study (1) the critical theory of phase transition between trivial and non-trivial SPTs, and (2) the relation between such critical theory and the gapless boundary theory of SPTs. Based on examples of SO(3)$\mathrm{SO}(3)$ and SU(2)$\mathrm{SU}(2)$ SPTs, we propose that under appropriate boundary condition the critical theory contains the delocalized version of the boundary excitations. In addition, we prove that the boundary theory is the critical theory spatially confined between two SPTs. We expect these conclusions to hold in general and, in particular, for discrete symmetry groups as well.     

基于Logistic回归建模和马尔可夫链蒙特卡罗方法计算后验描述丁酸梭菌株对于给定辐照剂量区的应答趋势

利用马尔可夫链蒙特卡罗（Markov Chain Monte Carlo,MCMC）方法估计Logistic回归模型中的参数,就是要构造一个以参数的后验分布为其平稳分布的非周期不可约的马尔可夫链,然后用该平稳分布中抽出的样本点计算蒙特卡罗积分。上述理论方法可以解决实验样本数据由于存在定和约束和多重共线性、在进行经典的logistic回归建模时的困难问题。基于此方法,研究了丁酸梭菌株对于给定辐照区间剂量的应答趋势,用模型挖掘数据所隐含的内在信息并导出了Logistic回归模型参数的贝叶斯框架下的50%,90%,95%和99%的置信区间。结果表明,运用Logistic与马尔可夫链耦合模型在有关给定辐射剂量对于微生物作用效果问题的logistic回归建模中具有较大的科学性与很好的使用性,从而可以为辐照诱变处理微生物制定辐照剂量区提供理论支持和回归技术借鉴。Using the Markov Chain Monte-Carlo method to estimate the parameters in the Logistic regression model, we constructed a non-periodic irreducible Markov Chain with the posterior distribution of the parameters as stationary distribution, and then used the sample points extracted from the stationary distribution to calculate the Monte-Carlo integral. The above theoretical method can solve the difficult problem of classical logistic regression modeling because of the existence and limitation of the experimental sample data and the multicollinearity. In the classical regression setup with a continuous response, the predicted values can range over all real numbers. Therefore, a different modelling technique is needed. In this work, the results describe in detail a previously unknown lethality trend following 12C6+ heavy-ion irradiation of Clostridium tyrobutyricum. By Markov Chain Monte-Carlo can calculate the model fit for a randomly selected subset of the chain and calculate the predictive envelope of the model. The grey areas in the plot correspond to 50%, 90%, 95%, and 99% posterior regions. More importantly, although this study focused on the use of the method in heavy-ion irradiation of microbial, its results are broadly applicable.