Séminaire CIMMUL | Li-Pang Chen

Statistical Machine Learning Methods for Noisy Survival Data Analysis

Li-Pang Chen

Université nationale Chengchi

Résumé

In medical studies and bioinformatics, an important research direction is the analysis of time-to-event data, where the main challenges often arise from incompleteness due to censoring mechanisms. With the growing ease of data collection, it is now common to encounter datasets with a large number of variables. Among these, even rare variables may carry valuable information. Another major challenge is measurement error, a typical feature of noisy data. In my presentation, I will introduce my recent work on survival analysis with multivariate or high-dimensional error-prone variables from the perspective of statistical machine learning. Specifically, I will first present graphical proportional hazards models, which incorporate network structures among variables. To simultaneously handle variable selection and network detection, I propose a penalized likelihood approach with a double-penalty function. Next, I will introduce the accelerated failure time model…