| Description |
Although the studies of causal inference and machine learning arose independently, there are emerging intersections which are proving fruitful for both fields. One of the major modern challenges in AI is to develop robust, generalizable models which can perform across a number of different tasks with minimal need for re-training and exposure to new data. Developing such models suggests a need for inference beyond prediction in settings with independent identically distributed data. Rather, the model should be able to leverage causal information in settings where interventions on a system can change the joint distribution of the data and classical statistical guarantees no longer apply. Causal inference has thus become a useful framework to ground research towards models capable of generalizing between tasks and answering questions about interventions on systems. In the other direction, much of the research in causal inference assumes knowledge of causal variables, and so machine learning is also becoming useful to causal inference for extracting high level causal information from low level data. In this paper, I begin by reviewing the foundational assumptions and concepts of causal inference, with an emphasis on those that relate to machine learning. I will cover open problems in machine learning, and how incorporating ideas from causal inference has led to progress in these areas. Then, I will discuss some machine learning approaches to answering causal questions, and perform experiments regarding causal effect estimation. Finally, I will finish with a discussion of future avenues for research in the intersection of causal inference and machine learning. |
