報告題目：Neural frailty machine: beyond proportional hazards assumption in neural survival regressions
報 告 人：郁文 教授 複旦大學
報告時間：2022年7月11日 14：00-15：00
報告地點：騰訊會議 ID：367231583 會議密碼：0711
校内聯系人：王培潔 wangpeijie@jlu.edu.cn
報告摘要：The Cox proportional hazards model is the most widely used regression model for survival data with right censoring. When the proportional hazards assumption does not hold, many alternative semiparametric models are proposed, including additive hazards models, accelerated failure time models, and linear transformation models, etc. Another way to extend the Cox model is to introduce frailty into the hazard function. Meanwhile, with recent development in machine learning, neural networks with deep structures are incorporated into the survival models to extend the linear structure. This brings great flexibility for survival data modeling. We consider a class of nonparametric hazard models with frailty and use neural networks to fit the model, which is called neural frailty machine. The likelihood function for right censoring is used to be the objective function. Two strategies are considered. One is to combine the non-parametric MLE and the neural networks and the other is to use the neural networks for all the function approximations. We show that the proposed estimators are consistent. In the experiments, we find out that the second strategy gives out better prediction performances more often, especially for data with large sizes.
報告人簡介：郁文，複旦大學管理學院統計與數據科學系教授、博士生導師、系主任。主要從事生存分析、半參數模型、兩階段抽樣設計、半監督推斷等研究，在國内外學術期刊發表論文三十篇，主持多項國家自然科學基金、教育部博士點基金研究工作。擔任中國現場統計研究會、全國工業統計學教學研究會、上海市質量技術應用統計學會理事。
