Causal feature selection for out-of-distribution generalization
Causal inference has been introduced to move beyond simple correlations, offering capabilities for explainability, generalization and action.
In this talk, I will focus on its use for generalization out-of-distribution (OOD), and how to leverage the causal structure to make predictions in a new environment through the causal invariance concept. We particularly propose a causal feature selection for OOD with latent confounders.
In this talk, I will focus on its use for generalization out-of-distribution (OOD), and how to leverage the causal structure to make predictions in a new environment through the causal invariance concept. We particularly propose a causal feature selection for OOD with latent confounders.
This work is in collaboration with Théotime Le Goff (PhD student).