Yushu Pan

We bridge the gap between causal image editing and large-scale text-to-image generation.

We develop a framework for building causally interpretable models where counterfactual queries can be consistently evaluated from a model and observational data.