时间：2018-12-21 09:30
地点：X2511，犀浦校区，第二教学楼
报告人：许晓菲，杭卫强，曾嘉杰
学术报告一： Regularized Partially Functional AutoRegressive Model in High Dimension with Application to Natural Gas Flows in Germany
摘要： We proposea partially functional autoregressive model (pFAR) to describe the dynamics of serial correlated functional data on both its own lagged value and high-dimensional exogenous scalar covariates. We develop a least square estimator under two-layer sparsity assumption, in particular, sparse group lasso penalty is imposed on functional and scalar covariates to select both important groups and important within-group elements simultaneously, which enables a detection of dependence structure and affecting exogenous factors. We establish the consistency properties of the proposed model under certain conditions, investigate the finite sample performance with simulation studies. In empirical analysis, we apply the pFAR model to natural gas data in Germany energy market and analysis how day-ahead natural gas is affected by some meteorological and economic factors and provides day-ahead forecast of the high resolution gas flow simultaneously.
报告人介绍： 许晓菲，本科毕业于中国科学技术大学，现在是新加坡国立大学统计与应用概率系的在读博士，导师为Chen Ying 副教授。许晓菲的主要研究方向为函数型时间序列的建模和预测及其在能源数据中的应用。
学术报告二： A Marginal Approach to Regressions with Additive Structures
摘要： For analysis of high dimensional data, combination of marginal regressions is an effective approach (Li et al, 2015; Fan et al, 2016). In this paper, we propose a boosting approach to the combina- tion of marginal regressions. We show that this procedure converges to the corresponding high dimensional regression models such as the addi- tive model and the varying coefficient model depending on the choice of marginal regressions. Asymptotic theory about the estimation procedure is established. This procedure can also incorporate other methods to deal with more complicated data. Simulation and real data analysis are used to demonstrate the usefulness of the proposed method.
报告人介绍： 杭卫强，本科毕业于电子科技大学，现在是新加坡国立大学统计与应用概率系的在读博士，导师为夏应存教授。杭卫强的主要研究方向为高维数据的统计分析以及高维数据的处理技术及其应用。
学术报告三：Computational methods for multi-target tracking
摘要：Bayesian inference methodology has been applied to the target tracking problem framework under application fields such as radar/sonar tracking, computer vision and robotics. The relevant theory and methods of Kalman filter for linear Gaussian dynamics and particle filter for more general scenario for single target tracking has been appropriately understood. As for the multiple targets’system involved with unknown association with observations, misdetections, and false alarms, more techniques have been introduced among the recent twenty years. These techniques can categorize into two types: I) filtering methods based on point process theory to model the varying number of targets which recursively predict and update the estimation each time step. These methods have solid theory foundation and its strength and drawback are well-studied. II) smoothing methods utilize MCMC to sample from the posterior distribution of targets states and data association with observations and targets. These methods often involve carefully designed MCMC propose move to explore the state space. In this talk, I will give a brief overview on these techniques available to solve a general class of tracking problem, and I will also mention some of my current work in this field at the end.
报告人介绍：曾嘉杰，本科毕业于电子科技大学，现在是新加坡国立大学统计与应用概率系的在读博士，导师为Ajay Jasra 教授。曾嘉杰主要从事基于卡尔曼滤波、马尔科夫链蒙特卡洛等方法的关于目标追踪的统计算法研究。
