Speaker: Dr. PANG Zhen, The Hong Kong Polytechnic University
 
Time: Dec. 20, 2016 (Tuesday) 16:00-17:00
 
Venue: 维格堂 319
 
Abstract:
A popular approach to model high dimensional data is to use the regularisation penalty. However, the final selected model is determined by a tuning parameter which controls the level of penalty. Thus, it is a fundamental problem to determine this tuning parameter. We propose a novel tuning parameter selection method by combining information criteria and stability selection to enjoy the advantages of the two methods while avoiding their drawbacks. The new method is quite general and can be generalised to many classes of models like the generalised linear models, Cox's proportional hazard model in survival analysis, graphical models and clustering.