作者：DMIR 来自： 发表时间：2019-07-01 浏览量：1093
报告题目：Causal Discovery and Prediction in the Presence of Distribution Shifts
报告人：卡内基梅隆大学Biwei Huang (黄碧薇)
Biwei Huang is a Ph.D. candidate in the philosophy department at Carnegie Mellon University (CMU), supervised by Prof. Kun Zhang and Prof. Clark Glymour. Before joining CMU, she studied at the University of Tübingen and Max-Planck Institute for Intelligent Systems in Germany. Her main research interests include causal discovery, machine learning, and computational neuroscience. She is actively exploring theoretical implementations in causal discovery, how causal knowledge facilitates learning problems, and practical uses of causality in neuroscience, biology, etc.
Many tasks in empirical sciences or engineering rely on the underlying causal information. As it is often difficult to carry out randomized experiments, inferring causal relations from purely observational data, known as the task of causal discovery, has drawn much attention. Over the last few years, with the rapid accumulation of huge volumes of data, causal discovery is facing exciting opportunities but also great challenges. One feature such data often exhibit is distribution shift. In this talk, I will present a principled framework for causal discovery from such data, called Constraint-based causal Discovery from heterogeneous/NOnstationary Data (CD-NOD). I will then show how to efficiently estimate the “driving force” of changing causal mechanisms after learning the causal structure.
On the other hand, due to the ubiquity of distribution shift, transfer learning has been an important research topic. In the second part of the talk, I will present an approach to time-varying causal modeling and prediction, where the causal coefficients follow dynamic models. Given the causal model, we treat prediction as a problem in Bayesian inference, which exploits the time-varying property of the data and adapts to new observations in a principled manner. Essentially, causal knowledge describes the changeability of the distribution and facilitates transfer learning.