ACTIVA: Amortized Causal Effect Estimation via Transformer-based Variational Autoencoder

ACTIVA is a transformer-based conditional variational autoencoder designed for amortized causal effect estimation from observational data. It introduces a conditional latent prior enabling zero-shot inference and demonstrates superior performance on synthetic datasets and gene-expression simulations compared to correlational and other amortized baselines. This model addresses the challenges of causal ambiguity and restrictive assumptions, making it relevant for practitioners focused on causal inference and interventional distribution estimation in AI applications.