|
Abstract:
|
There is growing recognition in the machine learning community that understanding causal relationships is crucial for solving a variety of pertinent problems, including policy design, sensitivity analysis, transfer learning, and fairness. Latent variables, however, can complicate causal inference by inducing problematic, spurious associations. In this talk, I will describe two techniques for negating such problematic associations in linear models, called Auxiliary Variables (AVs) and PushForward, and describe how they can be used to address the aforementioned problems.
|
