9月7日(周一)
模型不确定性下的状态估计:博弈论和数据驱动方法
时间:9:30
地点:正阳楼3号楼102
主讲人:Mattia Zorzi 帕多瓦大学信息工程学院 副教授
主讲人简介:Mattia Zorzi received the M.S. degree in Automation Engineering and the Ph.D. degree in Information Engineering from the University of Padova, Padova, Italy, in 2009 and 2013, respectively. He held postdoctoral appointments with the Department of Electrical Engineering and Computer Science, University of Liege, Liege, Belgium, and with the Human Inspired Technology Research Centre, University of Padova, Padova, Italy. He held visiting positions with the Department of Electrical and Computer Engineering, University of California, Davis, USA, and with the Department of Engineering, University of Cambridge, Cambridge, U.K., in 2011 and 2013-2014, respectively. He is currently an Associate Professor with the Department of Information Engineering, University of Padova. His current research interests include machine learning, deep learning, robust estimation, identification theory.
摘要:经典的卡尔曼滤波,虽已广泛应用于导航、机器人、金融以及环境监测等领域,然而该方法通常依赖于精确的状态空间模型信息。事实上,在许多实际场景中,系统模型的信息往往只能近似获得,并且不可避免地受到各种不确定性因素的影响。显然,在这种情况下,经典的滤波方法会使估计性能有所下降,甚至出现系统不稳定现象。本次报告将介绍一种基于极小极大博弈论方法的鲁棒状态估计框架。该方法将模型不确定性视为一个“对抗参与者”,并通过设计估计器,使其在最不利情形下的估计误差最小化,从而提高状态估计对模型不确定性的鲁棒性。随后,报告将进一步讨论该方法向非线性系统的推广。最后,报告还将探讨如何利用观测数据学习系统的不确定性信息,进而建立针对最坏情形下的鲁棒设计与数据驱动自适应方法之间的联系。
9月9日(周三)
报告题目一:双链协同建模的基因组语言模型
时间:9:00
地点:17幢401
主讲人:刘元盛 湖南大学 副教授
主讲人简介:刘元盛,湖南大学计算机学院副教授,博士毕业于澳大利亚悉尼科技大学。获湖南省自然科学奖一等奖(第二),ACM SIGBIO中国新星奖,入选湖南省人才项目湖湘青年英才。近年来以第一或通讯作者身份在 Nature Machine Intelligence、Bioinformatics、EMNLP 重要学术期刊和会议上发表学术论文40余篇。主持国家自然科学基金面上和青年项目。担任国际SCI期刊 Journal of Translational Medicine和BMC Biology 副编辑。
摘要:基因组语言模型通过在海量DNA序列上进行预训练,能够学习到序列的语法与语义规律,进而为调控元件识别、变异效应预测、序列设计等下游任务提供通用表征。然而,现有gLM大多将DNA视为静态单链序列,忽略了双链之间在转录、复制与表观遗传等关键过程中的动态信息交互。本报告将介绍我们提出的基因组语言模型——CrossDNA。该模型在架构层面显式建模DNA双链的动态交互机制,通过双分支交叉视图输入、链间信息通信、Comba与滑动窗口注意力协同实现长程上下文建模,并结合自蒸馏约束提升表征稳定性。CrossDNA在调控元件分类、染色质图谱预测、增强子独立测试、跨物种泛化以及长序列eQTL预测等多项任务中均取得领先性能,展现出更强的方向鲁棒性、参数效率与零样本表征能力。该工作为从静态序列近似走向动态生物过程建模提供了新路径。
报告题目二:A second-order generalization of TC and DC kernels
时间:9:30
地点:正阳楼3号楼202
主讲人:Mattia Zorzi
摘要:Kernel-based methods have been successfully introduced in system identification to estimate the impulse response of a linear system. Adopting the Bayesian viewpoint, the impulse response is modeled as a zero mean Gaussian process whose covariance function (kernel) isestimated from the data. The most popular kernels used in system identification are the tuned-correlated (TC), the diagonal-correlated (DC) and the stable spline (SS) kernel. TC and DC kernels admit a closed-form factorization of the inverse. The SS kernel induces more smoothness than TC and DC on the estimated impulse response, however, the aforementioned property does not hold in this case. In this talk we propose a second-order extension of the TC and DC kernel, which induces more smoothness than TC and DC, respectively, on the impulse response and ageneralized-correlated kernel, which incorporates the TC and DC kernels and their second order extensions. Moreover, these generalizations admit a closed-form factorization of the inverse and thus they allow to design efficient algorithms for the search of the optimal kernel hyperparameters. We also show how to use this idea to develop higher oder extensions. Interestingly, these new kernels belong to the family of the so called exponentially convex local stationary kernels: such a property allows to immediately analyze the frequency properties induced on the estimated impulse response by these kernels.
报告题目三:Identification of forward models: a nonparametric approach
时间:14:00
地点:正阳楼3号楼202
主讲人:Mattia Zorzi
摘要:In this talk, we present a new kernel-based method for identifying the impulse responses of forward (or simulation) models from input-output data. While traditional regularized methods re-parameterize systems via one-step ahead predictors, they make it remarkably difficult to encode crucial prior information -- such as stability -- directly into the forward model. To overcome this limitation, we frame the problem directly in terms of the forward model's impulse responses, leading to a nonlinear, infinite-dimensional Tikhonov regularization problem. We prove the existence of a solution, generalize the classical representer theorem to characterize its structure, and show that this justifies approximating the forward system via a high-order MAX (Moving Average with eXogenous input) model. Lastly, we address the problem to tune the kernel hyperparameters from data.
9月11日(周五)
“人之为人”与“连接的中枢”: AI时代重塑人文学
时间:10:00-
地点:外国语学院22幢301
主讲人:颜海平 清华大学清华学堂世文首席教授
主讲人简介:颜海平,清华大学清华学堂世文首席教授、外文系与中文系教授博导、清华大学世界文学与文化研究院院长、清华大学校务委员会委员、国务院第七届外国语言文学学科评议组成员、教育部清华大学中外人文交流研究中心主任、中英高等教育人文联盟执行理事会理事长兼秘书长。2014-2020年任清华大学外国语言文学系主任、2015-2025清华大学校学术委员会委员。毕业于复旦大学,后于康奈尔大学获文学与思想史、批判哲学、跨文化理论博士。在美执教二十年,先后任加州大学洛杉矶分校和康奈尔大学终身资深教授。2012年特聘国际人文学专家全时回国。现任康奈尔大学Diacritics学术指导委员会终身成员、普林斯顿大学出版社中国学术顾问理事会理事。主要聚焦跨国族女性主义、跨文化现代主义、批判性世界主义、复语文学与研究等;获学术奖项、荣誉三十余项。代表著述与论文有《中国现代女性作家与中国革命》《布莱希特与中国古典剧场性》《从“欧洲”到“西方”的时空之变》《另一种世界主义》《连接的中枢》《互为的转写》《全球迁徙者的登场》等。任商务印书馆Cosmopolitics国际资深学者专著丛书作者与主编;英国劳特利奇出版社“Transcultural Studies and China (跨文化研究与中国)”专著与文辑系列作者与主编;三联书店“跨国别与区域研究译丛”导读作者与主编;商务印书馆“重塑人类思想史”世界名著译丛总主编;商务印书馆《重塑人文学:中英对话》总主编。早年本科期间发表十幕大型历史剧《秦王李世民》获文化部与中国剧协授予的1981-1982年全国优秀剧本一等奖。该剧收入2009年《曹禺剧本奖作品选》。
编辑:武艳