High-Dimensional Data Analysis Using Greedy Algorithms

Speaker：

Ching-Kang Ing (NTHU)

Time：

Thur.,13:00-16:00

Oct. 17 / 24, 2024

Venue：

Jingzhai 105

Description:

Greedy algorithms, particularly the orthogonal greedy algorithm (OGA), are frequently used for high-dimensional model selection as an alternative to Lasso. In this course, I will discuss the statistical properties of OGA when applied in conjunction with high-dimensional criteria for model selection in both stationary and non-stationary high-dimensional time series models. Additionally, I will explain how greedy-type algorithms can be employed to estimate high-dimensional sparse covariance matrices of stationary time series. Furthermore, I will demonstrate modifications of OGA for high-dimensional model selection in the presence of covariate shift. If time permits, I will introduce the performance of the Chebyshev Greedy Algorithm (CGA), a non-linear counterpart of OGA, in certain high-dimensional non-linear models.