报 告 人：韦基球数据科学讲座教授及香港岭南大学校长

报告时间：6月25日（周二）15:30-16:45

参会地点：伯川图书馆多功能厅201A

报告摘要：

In this talk I will present a novel latent vector autoregressive framework to model reduced dimensional dynamic components. High dimensional time series data are common in modern engineering, internet of things, financial, economic, autonomous, and control systems. Dimension reduction and dynamics modeling tasks are simultaneously required, but traditional multivariate time series analysis does not handle them simultaneously. We present a probabilistic reduced-dimensional vector autoregressive (PredVAR) model to extract low-dimensional dynamics with a canonical correlation analysis (CCA) objective. The dynamic latent variable scores are enforced with a reduced dimensional VAR model with maximized predictability. The model utilizes an oblique projection to partition the measurement space into a subspace that contains the reduced-dimensional dynamics and a complementary static subspace. We develop an iterative PredVAR algorithm using maximum likelihood and the expectation-maximization (EM) framework. Time-dependent data from a chaotic Lorenz oscillator and an industrial process are used to test the superiority of the proposed algorithm. The reduced-dimensional latent dynamic modeling framework has potentially wide applications in prediction, control, and diagnosis of anomalies.

个人简介：

秦泗钊- 香港岭南大学校长
获清华大学自动化学士，硕士，马里兰大学化学工程博士。曾任美国得克萨斯大学Austin 分校及南加州大学教授25年，2011-2013年兼任南加州大学工学院副院长。2020年1月起回香港担任香港城市大学数据科学学院首任院长。研究领域包括工业大数据，自动化，工业智能。入选美国国家发明科学院院士，欧洲科学与艺术院院士，香港工程院院士。学术荣誉包括美国基金会事业成就奖，美国化工学会CAST化工计算奖，IEEE CSS 技术转化奖，IEEE 会士，美国化工学会会士，国际自动控制联盟 （IFAC）会士，中国教育部长江学者等。