Performance comparison of auto-tuned causal inference model architectures
The library for causal inference AutoML, https://github.com/transferwise/auto-causality, has now integrated models from the EconML library that use the Double Machine Learning, Doubly Robust Learning, and OrthoForest methods, for both unconfounded CATE and instrumental variable problems. This thesis would integrate the DeepIV model from EconML as well as several other neural network approaches from the literature, and compare their performance, including hyperparameter tuning, to those of the already available models, using benchmark datasets. As an optional extension, automated neural architecture search could be used to further refine those neural network approaches.
Lõputöö kaitsmise aasta
Egor Kraev (Wise), Prof. Radwa El Shawi