A fundamental causal modelling task is to predict the effect of an intervention (or treatment) on an outcome, given context/covariates. Examples include predicting the effect of a medical treatment on patient health given patient symptoms and demographic information, or predicting the effect of ticket pricing on airline sales given seasonal fluctuations in demand. The problem becomes especially challenging when the treatment and context are complex (for instance, “treatment” might be a web ad design or a radiotherapy plan), and when only observational data is available (i.e., we have access to historical data, but cannot intervene ourselves). The challenge is greater still when the covariates are not observed, and constitute a hidden source of confounding. 

I will give an overview of some practical tools and methods for estimating causal effects of complex, high dimensional treatments from observational data. The approach is based on conditional feature means, which represent conditional expectations of relevant model features. These features can be deep neural nets (adaptive, finite dimensional, learned from data), or kernel features (fixed, infinite dimensional, enforcing smoothness). When hidden confounding is present, a neural net implementation of instrumental variable regression can be used to correct for this confounding. The methods will be applied to modelling employment outcomes for the US Job Corps program for Disadvantaged Youth, and in policy evaluation for reinforcement learning.
 
Further reading: For observed confounders, Kernel Methods for Causal Functions: Dose, Heterogeneous, and Incremental Response Curves (Biometrika 2023), A Neural Mean Embedding Approach for Back-door and Front-door Adjustment (ICLR 2023). 
 
For hidden confounders, Kernel Instrumental Variable Regression (NeurIPS 2019), Learning Deep Features in Instrumental Variable Regression (ICLR 2021), Proximal Causal Learning with Kernels: Two-Stage Estimation and Moment Restriction (ICML 2021), Deep Proxy Causal Learning and its Application to Confounded Bandit Policy Evaluation (NeurIPS 2021).

Bio: Arthur Gretton is a Professor with the Gatsby Computational Neuroscience Unit; director of the Centre for Computational Statistics and Machine Learning (CSML) at UCL; and Research Scientist at Google Deepmind. He received degrees in Physics and Systems Engineering from the Australian National University, and a PhD with Microsoft Research and the Signal Processing and Communications Laboratory at the University of Cambridge. He previously worked at the MPI for Biological Cybernetics, and at the Machine Learning Department, Carnegie Mellon University.
Arthur’s recent research interests in machine learning include causal inference and representation learning, design and training of generative models (implicit: Wasserstein gradient flows, GANs; and explicit: energy-based models), and nonparametric hypothesis testing.
He has been an associate editor at IEEE Transactions on Pattern Analysis and Machine Intelligence from 2009 to 2013, an Action Editor for JMLR since April 2013, an Area Chair for NeurIPS in 2008 and 2009, a Senior Area Chair for NeurIPS in 2018 and 2021, an Area Chair for ICML in 2011 and 2012, a Senior Area Chair for ICML in 2022, a member of the COLT Program Committee in 2013, and a member of Royal Statistical Society Research Section Committee since January 2020. Arthur was program chair for AISTATS in 2016 (with Christian Robert), tutorials chair for ICML 2018 (with Ruslan Salakhutdinov), workshops chair for ICML 2019 (with Honglak Lee), program chair for the Dali workshop in 2019 (with Krikamol Muandet and Shakir Mohammed), and co-organsier of the Machine Learning Summer School 2019 in London (with Marc Deisenroth).