Sonali Parbhoo - Information Bottleneck Methods for Estimating Treatment Effects [IndabaX SA 2020]
Автор: Deep Learning IndabaX
Загружено: 2020-03-26
Просмотров: 464
Talk by Sonali Parbhoo from Harvard University at the Deep Learning Indaba𝕏 South Africa 2020
[https://indabax.co.za]
Estimating the effects of an intervention from high-dimensional observational data is a challenging problem due to the presence of confounding. In applications such as healthcare the task is often further complicated where certain observations are systematically missing at test time, thereby prohibiting accurate inference for new patients. In this talk, we propose an approach based on the information bottleneck principle to reason about the effects of interventions. To this end, we first use the information bottleneck principle to perform a low-dimensional compression of covariates by explicitly considering the relevance of information for treatment effects. As a second step, we subsequently use the sufficiently reduced covariate to perform a transfer of the relevant information to cases where data are missing at test time, allowing us to reliably and accurately estimate treatment effects even where data is incomplete. Our results on two causal inference benchmarks and a real application for treating sepsis show that our method achieves state-of-the-art performance, without compromising interpretability.
Доступные форматы для скачивания:
Скачать видео mp4
-
Информация по загрузке: