Event
Seminar: Estimating Causal Effect – Challenges and Countermeasures
Causal inference is increasingly important in guiding decision-making in high-stake domains, such as healthcare, education, e-commerce, etc.
Please join us for CIRES HDR, Hechuan Wen’s, PhD confirmation milestone seminar.
Estimating Causal Effect: Challenges and Countermeasures
Speaker: Hechuan Wen
Abstract: Causal inference is increasingly important in guiding decision-making in high-stake domains, such as healthcare, education, e-commerce, etc. Normally, randomized control trial (RCT) is the gold standard for estimating the causal effect. Given that implementing RCTs is costly, time-consuming, and sometimes ethically intractable, various applications alternatively turn to using the passively collected observational data to perform causal inference in a data-driven fashion.
To date, various neural methods have been proposed for causal effect estimation based on the independent observational data (non-graph), where a default assumption is the same distribution and availability of variables at both training and inference (i.e., runtime) stages. However, distribution shift (i.e., domain shift) could happen during runtime, and bigger challenges arise from the impaired accessibility of variables. We term the co-occurrence of domain shift and inaccessible variables runtime domain corruption, which seriously impairs the generalizability of a trained counterfactual predictor.
Recently, researchers extended the causal effect estimation to the networked data (graph), where richer information can be leveraged to better help deconfound. However, little attention has been paid to the core assumption in causal effect estimation – the Positivity assumption. If such an assumption is indeed violated, estimating the causal effect can express high error. In such circumstances, identifying the possibly high error estimation is crucial since a wrong decision based on the estimated effects can result a consequence.
To deal with the aforementioned two unsolved challenges respectively: 1) we propose a two-stage domain adaptation scheme to handle the domain corruption issue on the independent data; 2) we propose an uncertainty quantification method to discover the estimations which tend to express high error on the networked data. Extensive experiments demonstrate the superiority of the proposed methods.
Speaker: Hechuan Wen completed his B.Eng. in Aircraft Engineering at Nanjing University of Aeronautics and Astronautics in 2019, and his M.Com. in Business Analytics at The University of Sydney in 2021. Currently, he is a PhD student at the School of Information Technology and Electrical Engineering, the University of Queensland, under the supervision of Dr. Rocky Chen, A/Prof. Hongzhi Yin and Prof. Shazia Sadiq. His research interests include causal inference (under potential outcome framework), uncertainty quantification and variance reduction.
View all events