Double Descent

We study the problem of learning causal models from observational data through the lens of interpolation and its counterpart---regularization. A large volume of recent theoretical, as well as empirical work, suggests that, in highly complex model …