École Nationale Supérieure de
Techniques Avancées

Résumé

The goal of this internship is to explore causal discovery in the presence of biased data. To achieve this, we first developed a synthetic data generator based on a causal graph (the ground truth). This generator takes a causal graph and a specified bias level as inputs, and produces synthetic data that maintains the same causal structure as the ground truth while incorporating the given level of bias. Using this synthetic data generator, we create datasets with varying levels of bias for each ground truth. We then apply causal discovery algorithms to these datasets to investigate how the presence and extent of bias affect the accuracy and structure of the discovered causal relationships.