Xiaochuan Li

I study when a model's stable behavior is actually licensed by the causal structure we think it is using. My work asks when model behavior can be causally reduced to the minimum structure that licenses it.