Mouad EL Bouchattaoui
The thesis presents novel methods for causal inference and representation learning in high-dimensional, time-varying data, introducing models like CDVAE and frameworks utilizing RNNs and CPC for improved causal analysis.
This research focuses on improving how we understand cause-and-effect relationships in complex, changing data over time. It introduces new models and techniques to better estimate how different treatments affect individuals by capturing hidden factors that influence outcomes. One key model, CDVAE, is shown to perform better than existing methods and can enhance other models significantly. Additionally, the study presents a framework using RNNs to predict long-term outcomes more efficiently, and it offers new ways to interpret how hidden causes affect observable data features.
