Adversarial Generalized Method of Moments
This provides a method for social scientists to improve causal inference models, but it is incremental as it adapts existing adversarial techniques like GANs to a specific domain.
The paper tackles the problem of estimating models described by conditional moment restrictions, which are crucial for causal inference in social sciences, by formulating it as a zero-sum game between a modeler and an adversary and applying adversarial training, achieving practical performance in non-parametric instrumental variable regression.
We provide an approach for learning deep neural net representations of models described via conditional moment restrictions. Conditional moment restrictions are widely used, as they are the language by which social scientists describe the assumptions they make to enable causal inference. We formulate the problem of estimating the underling model as a zero-sum game between a modeler and an adversary and apply adversarial training. Our approach is similar in nature to Generative Adversarial Networks (GAN), though here the modeler is learning a representation of a function that satisfies a continuum of moment conditions and the adversary is identifying violating moments. We outline ways of constructing effective adversaries in practice, including kernels centered by k-means clustering, and random forests. We examine the practical performance of our approach in the setting of non-parametric instrumental variable regression.