Abstract:
This work is a review on causal inference with a focus on predicting temporal Individual Treatment Effects (ITEs). Various approaches were investigated, and the Causal Effect Variational Autoencoder (CEVAE) was selected as the most promising method for this purpose. In many real-life scenarios, there is a pressing need to understand not just the static causal relationships between variables, but also how these relationships evolve and interact over time. The Temporal Causal Effect Variational Autoencoder (T-CEVAE) is particularly motivated by such settings where time-varying confounders play a critical role. In our research, we’re using a tool called Causal Effect Variational Autoencoders (CEVAE), teamed up with a memory-friendly system called Long Short-Term Memory (LSTM). This helps us understand the time-related patterns in data better. The project summarizes the findings on causal inference and discusses the potential of CEVAE in predicting temporal ITEs.