T61: Nonlinear causal discovery with additive noise models
Patrik Hoyer, Dominik Janzing, Joris Mooij, Jonas Peters, Bernhard Schölkopf

New approach to infer whether x causes y or y causes x from statistical data Assumption: the effect is a function of the cause up to an additive noise term that is statistically independent of the cause

** * * * * * ** effect * *** * * * * * * cause * * * *

f(cause)

Theorem: in the generic case, the causal direction is identifiable because there won't be such a model from the effect to the cause Experiments: correct for 8 out of 9 data sets with known ground truth