题目： Sampling based inference for logistic regression

报告人：HaiYing Wang教授（University of Connecticut）

Abstract : 

In this talk, we first introduce an optimal subsampling approach in the context of logistic regression, where the subsampling probabilities are derived to minimize the asymptotic mean squared error of the subsample estimator. Next, we propose an improved estimation method based on unweighted subsample target function with bias correction. Pilot estimators are required to calculate optimal subsampling probabilities and to correct biases; interestingly, even if pilot estimators are inconsistent, the proposed method still produce consistent and asymptotically normal estimators. We also develop a new algorithm based on Poisson subsampling, which does not require to use the subsampling probabilities all at once and has a higher estimation efficiency. Lastly, we consider the case of big data with rare events. We derive asymptotic distributions to demonstrate how many data do we really have in this scenario, and show that a subsample estimator can be as efficient as the full data estimator. It is a common practice to over-sample the cases with rare events data. We will demonstrate why this approach is not recommended. Both theoretical and numerical results will be presented.

时间：2019年12月24日（周二）下午16:00-17:00地点：565net必赢客户端教二楼610教室

联系人：邹国华

欢迎全体师生积极参加！