Hang Wu

Machine Learning Thesis Proposal Presentation

Date: December 12th, 2019

Time: 9:00 am - 11:00 am

Location: Coda C1115 Druid Hills 

Committee Members:

May D. Wang, PhD (Georgia Tech/Emory, Department of Biomedical Engineering) (Advisor)

Polo Duen Horng Chau, PhD (Georgia Tech, School of Computational Science & Engineering)

Justin Romberg, PhD (Georgia Tech, Department of Electrical Engineering)

Title: Adaptive Causal Inference using Learning-to-Learn Techniques

Summary

Causal inference appears in a wide range of domains, for example, causal relationships between molecules, the causal effect of a public policy, building invariant machine learning models. However, the limited sample size and the heterogeneity of causal models make it challenging to apply causal inference to real-world applications. While humans excel in learning from a few samples and quickly adapt to unseen tasks, can we build causal inference algorithms that have similar efficiency and flexibility? 

This proposal outlines our previous and proposed work for developing adaptive causal inference algorithms using learning-to-learn techniques. First, we present adaptive causal effect estimation algorithms, and demonstrate its applications in clinical decision support and recommendation systems. Second, we propose algorithms for quickly identifying multiple correlated causal graphs using learning-to-learn principles. Lastly, we present applications of causal inference in fairness of machine learning.