Thread by @volokuleshov

How can deep learning be useful in causal inference?

In our #NeurIPS2022 paper, we argue that causal effect estimation can benefit from large amounts of unstructured "dark" data (images, sensor data) that can be leveraged via deep generative models to account for confounders.

Consider the task of estimating the effect of a medical treatment from observational data. The true effects are often confounded by unobserved factors (e.g., patient lifestyle). We argue that latent confounders can be discovered from unstructured data (e.g., clinical notes).

For example, suppose that we have access to raw data from wearable sensors for each patient. This data implicitly reveals whether each patient is active or sedentary—an important confounding factor affecting treatment and outcome. Thus, we can also correct for this confounder.

In our paper, we propose a deep generative model of treatments t, outcomes y, latent confounders z, and multiple types of unstructured data x_i. This model generalizes structural equations—when it fits the data well, it yields estimates of treatment effect.

Learning this deep latent generative model can be challenging. We provide a variational inference algorithm that leverages the graphical structure of the model, naturally deals with multiple sources of data possibly missing at random, and scales to large datasets.

We apply this model to a wide range of tasks, including genome-wide association studies in plants. We show that leveraging unstructured weather data can help more accurately identify the causal effect of genetic mutations.

This work was led by Cornell PhD student Shachi Deshpande. If you missed Shachi’s presentation at NeurIPS, check out our paper here: openreview.net/pdf?id=ByYFpTw…