In this talk, I will introduce a novel approach to robust time series prediction, which is formulated as a structured output regression problem by simultaneously considering the prediction error and the shape of the predicted sequence in model training. I also show that a simplified version of the generic model employing linear regression and PCA leads to a single optimization problem with convex objective function and globally optimal solution.

Add:Building 58, The School of Mathematics, Tianjin University Beiyangyuan Campus,

No. 135, Ya Guan Road, Jinnan District, Tianjin, PRC

Tel:022-60787827 Mail:math@tju.edu.cn