Yuchen Zhu gave a talk on September 20th, at 10 AM, in the Lecture Hall (ground floor) of the AI Research Building, Maria-von-Linden-str. 6, Tübingen

Full video

Abstract

Causal inference is necessary in many social science domains for understanding the effects of interventions such as that of a new teaching strategy. Despite its promise, using causal inference for decision-making is still at an early stage - among others, one common concern is the hardness of causal structure identification which appears to form the basis of many treatment estimation algorithms, and another being the low signal-to-noise ratio in social science data. In this talk, I will first discuss the characteristics of social science data and how we can circumvent them in causal inference. I will then follow this by introducing several specific modern Causal Inference methodologies. For each methodology I introduce, I will pay particular attention to discussing their assumptions with the aim of drawing connections with the problem settings in social sciences.

Bio

Yuchen is a second-year PhD student at UCL and currently a research intern at Amazon Science Tübingen. She is particularly interested in causal machine learning methods applicable to social sciences. This leads her to explore flexible ways of causal effect estimation under weak conditions, where confounding is present and clean data satisfying causal identification conditions are not available. Additionally, due to the high-stake nature of many social science applications, she naturally finds herself working with methods which exhibit theoretical guarantees.