第三届丝绸之路青年学者研讨会数学与统计学院分论坛学术信息
第三届丝绸之路青年学者研讨会数学与统计学院分论坛学术信息
论坛时间：2017年11月17日（周五）下午
论坛地点：理科楼112会议室
讲座一：赵凯峰博士讲座
讲座题目：贝叶斯建模及有效算法
讲座摘要：近些年，贝叶斯方法在统计学及相关领域得到了长足的发展（如贝叶斯统计、概率图模型等），渐渐形成了一个重要的学科分支。其建模思路的一致性及相关算法的有效性使其在数据处理、分析和建模等领域得到了广泛的应用。本次报告将以报告人的科研成果为例，简述贝叶斯方法在统计学习领域中几个典型的应用实例。具体包括其在半参模型、分位数回归模型、变量选择、成分识别等热点问题中的应用。同时，此次报告也将介绍若干贝叶斯方法中常用的有效算法。
讲座人简介：赵凯峰，群邑数据科学全球研发中心，Research Scientist，2015年博士毕业于新加坡南洋理工大学统计学专业。毕业后加入全球顶尖广告&互联网公司，在数据科学部担任研发科学家，致力于数据科学方法研究及面向业界应用的科技成果转化，在相关领域共发表SCI论文十余篇，同时担任SCI期刊审稿人。
讲座二：王南伟博士讲座
讲座题目： High-dimensional Graphical Model Learning
讲座摘要：Graphical model is a classical statistical model for identifying the conditional independence relationship among random variables. Nowadays high throughput sequencing technology in genomic study brought new challenges in both graphical model learning and parameter estimation. In this talk, I will talk about my phd research in discrete graphical models and the application of graphical models in genome wide association study.
讲座人简介：王南伟，加拿大多伦多大学，博士后，2017年博士毕业于加拿大约克大学数学统计专业。目前从事生物统计的研究工作，主要研究方向是不同基因表示数据的相关性，以及它们对不同疾病的影响。博士期间主要的研究方向是graphical models，在annals of statistics、journal of multivariate analysis 各发表文章一篇。
讲座三：周宗政博士讲座
讲座题目：Phase Transitions of Face-cubic Model on the Complete Graph
讲座摘要：In this talk, I will give an example of how large deviation analysis can be applied to study phase transitions of statistical mechanical models. We focus only on the complete graph, on which making rigorous analysis is possible. The model we considered is called n-component face-cubic model, and the nature of its phase transitions on the complete graph is still an open question since introduced in 1975. In this work, using large deviation analysis, we prove limit theorems for the magnetization which reveals that the nature of phase transitions is continuous for n <= 3 and discontinuous for n > 3.
讲座人简介：周宗政，澳大利亚莫纳什大学，博士后，2016年博士毕业于澳大利亚莫纳什大学数学科学学院。先后于墨尔本大学数学与统计学院、莫纳什大学数学科学学院做博士后研究，主要研究方向为统计力学、马尔科夫链蒙特卡洛方法，以及应用概率统计等，在Physical Review Letters, Physical Review E 等核心期刊发表论文9篇，担任Physical Review E 审稿人。
讲座四：刘升恒博士讲座
讲座题目：Structure-Aware Time-Frequency Analysis Algorithms for Sparse Nonstationary Signals
讲座摘要：In this seminar, novel time-frequency analysis algorithms exploiting the a priori knowledge on the signal structures will be introduced. In particular, a numerical algorithm will be presented for the fast computation of large-scale discrete fractional Fourier transform when the signal spectrum can be sparsely represented in the fractional Fourier domain. Also, its application to the coherent integration of accelerating targets in passive bistatic radars will be addressed. In addition, a comprehensive scheme for robust astronomic signal reception in the presence of fast FH interference and missing data observations will be introduced. This novel scheme consists of a sparse Bayesian learning based frequency-hopping spectrum reconstruction stage and a sparse cubic phase function based weak linear frequency modulated signal detection stage. To demonstrate the superior performance of both proposed approaches, simulation and real-data experiment results will be provided.
讲座人简介：刘升恒，英国爱丁堡大学，博士后研究员，2017年于北京理工大学获工学博士学位，期间赴美国天普大学工程学院联合培养。2017年4月至今在爱丁堡大学从事博士后研究工作，主要研究方向为非平稳信号时频分析、外辐射源雷达信号处理、压缩感知与稀疏学习等。发表论文23篇，其中SCI 12篇、EI 11篇，获国家发明专利授权5项。担任国际顶级学术期刊IEEE TSP, IEEE TASLP, IEEE TIM等审稿人。
讲座五：姚经教授讲座
讲座题目：A Steins type Lemma for the Multivariate Generalized Hyperbolic Distribution
讲座摘要: Markowitz' mean-variance optimization framework is known to be consistent with expected utility maximization under the assumption that returns are multivariate elliptically distributed or when a quadratic utility function is employed. Both assumptions are questionable. By contrast, the multivariate generalized hyperbolic distribution is known to provide an excellent fit to returns. By extending Stein's (1973) seminal lemma to this context we are able to completely describe the optimal portfolio of an expected utility maximizer. In particular, he invests in three funds instead of two (as in mean-variance optimization).
讲座人简介：姚经，比利时布鲁塞尔自由大学，科研教授，2013年博士毕业于比利时布鲁塞尔自由大学应用经济学专业，现任比利时布鲁塞尔自由大学科研教授、比利时天主教鲁汶大学访问学者、以色列海法大学精算研究中心研究员。主要从事量化金融分析，精算理论和风险管理领域相关研究。目前已发表论文7篇专著一本，主持科研课题两项等。
论坛地点：理科楼112会议室
讲座一：赵凯峰博士讲座
讲座题目：贝叶斯建模及有效算法
讲座摘要：近些年，贝叶斯方法在统计学及相关领域得到了长足的发展（如贝叶斯统计、概率图模型等），渐渐形成了一个重要的学科分支。其建模思路的一致性及相关算法的有效性使其在数据处理、分析和建模等领域得到了广泛的应用。本次报告将以报告人的科研成果为例，简述贝叶斯方法在统计学习领域中几个典型的应用实例。具体包括其在半参模型、分位数回归模型、变量选择、成分识别等热点问题中的应用。同时，此次报告也将介绍若干贝叶斯方法中常用的有效算法。
讲座人简介：赵凯峰，群邑数据科学全球研发中心，Research Scientist，2015年博士毕业于新加坡南洋理工大学统计学专业。毕业后加入全球顶尖广告&互联网公司，在数据科学部担任研发科学家，致力于数据科学方法研究及面向业界应用的科技成果转化，在相关领域共发表SCI论文十余篇，同时担任SCI期刊审稿人。
讲座二：王南伟博士讲座
讲座题目： High-dimensional Graphical Model Learning
讲座摘要：Graphical model is a classical statistical model for identifying the conditional independence relationship among random variables. Nowadays high throughput sequencing technology in genomic study brought new challenges in both graphical model learning and parameter estimation. In this talk, I will talk about my phd research in discrete graphical models and the application of graphical models in genome wide association study.
讲座人简介：王南伟，加拿大多伦多大学，博士后，2017年博士毕业于加拿大约克大学数学统计专业。目前从事生物统计的研究工作，主要研究方向是不同基因表示数据的相关性，以及它们对不同疾病的影响。博士期间主要的研究方向是graphical models，在annals of statistics、journal of multivariate analysis 各发表文章一篇。
讲座三：周宗政博士讲座
讲座题目：Phase Transitions of Face-cubic Model on the Complete Graph
讲座摘要：In this talk, I will give an example of how large deviation analysis can be applied to study phase transitions of statistical mechanical models. We focus only on the complete graph, on which making rigorous analysis is possible. The model we considered is called n-component face-cubic model, and the nature of its phase transitions on the complete graph is still an open question since introduced in 1975. In this work, using large deviation analysis, we prove limit theorems for the magnetization which reveals that the nature of phase transitions is continuous for n <= 3 and discontinuous for n > 3.
讲座人简介：周宗政，澳大利亚莫纳什大学，博士后，2016年博士毕业于澳大利亚莫纳什大学数学科学学院。先后于墨尔本大学数学与统计学院、莫纳什大学数学科学学院做博士后研究，主要研究方向为统计力学、马尔科夫链蒙特卡洛方法，以及应用概率统计等，在Physical Review Letters, Physical Review E 等核心期刊发表论文9篇，担任Physical Review E 审稿人。
讲座四：刘升恒博士讲座
讲座题目：Structure-Aware Time-Frequency Analysis Algorithms for Sparse Nonstationary Signals
讲座摘要：In this seminar, novel time-frequency analysis algorithms exploiting the a priori knowledge on the signal structures will be introduced. In particular, a numerical algorithm will be presented for the fast computation of large-scale discrete fractional Fourier transform when the signal spectrum can be sparsely represented in the fractional Fourier domain. Also, its application to the coherent integration of accelerating targets in passive bistatic radars will be addressed. In addition, a comprehensive scheme for robust astronomic signal reception in the presence of fast FH interference and missing data observations will be introduced. This novel scheme consists of a sparse Bayesian learning based frequency-hopping spectrum reconstruction stage and a sparse cubic phase function based weak linear frequency modulated signal detection stage. To demonstrate the superior performance of both proposed approaches, simulation and real-data experiment results will be provided.
讲座人简介：刘升恒，英国爱丁堡大学，博士后研究员，2017年于北京理工大学获工学博士学位，期间赴美国天普大学工程学院联合培养。2017年4月至今在爱丁堡大学从事博士后研究工作，主要研究方向为非平稳信号时频分析、外辐射源雷达信号处理、压缩感知与稀疏学习等。发表论文23篇，其中SCI 12篇、EI 11篇，获国家发明专利授权5项。担任国际顶级学术期刊IEEE TSP, IEEE TASLP, IEEE TIM等审稿人。
讲座五：姚经教授讲座
讲座题目：A Steins type Lemma for the Multivariate Generalized Hyperbolic Distribution
讲座摘要: Markowitz' mean-variance optimization framework is known to be consistent with expected utility maximization under the assumption that returns are multivariate elliptically distributed or when a quadratic utility function is employed. Both assumptions are questionable. By contrast, the multivariate generalized hyperbolic distribution is known to provide an excellent fit to returns. By extending Stein's (1973) seminal lemma to this context we are able to completely describe the optimal portfolio of an expected utility maximizer. In particular, he invests in three funds instead of two (as in mean-variance optimization).
讲座人简介：姚经，比利时布鲁塞尔自由大学，科研教授，2013年博士毕业于比利时布鲁塞尔自由大学应用经济学专业，现任比利时布鲁塞尔自由大学科研教授、比利时天主教鲁汶大学访问学者、以色列海法大学精算研究中心研究员。主要从事量化金融分析，精算理论和风险管理领域相关研究。目前已发表论文7篇专著一本，主持科研课题两项等。