Learning Unkown Intervention Targets in Structural Causal Models
- Saber Salehkaleybar
- Friday 7 July 2023
Niels Bohrweg 1
2333 CA Leiden
Randomized interventions on a target variable are utilized to estimate the causal effect of the target. However, in some applications, we may not have full control in terms of which variables are intervened on. For instance, in recovering causal protein-signaling networks from single-cell data, drugs are injected into cells to inhibit or activate some signaling proteins, and gene expression levels are measured. In these experiments, the intervention targets are unknown.
In this talk, I first review some recent advances in the problem of nonlinear independent component analysis (ICA). Then, I present an application of ICA in the problem of identifying unknown intervention targets in structural causal models where we have access to heterogeneous data collected from multiple environments. The unknown intervention targets are the set of endogenous variables whose corresponding exogenous noises change across the environments. I discuss a two-phase method which in the first phase recovers the exogenous noises corresponding to unknown intervention targets whose distributions have changed across environments. In the second phase, the recovered noises are matched with the corresponding endogenous variables. Under the causal sufficiency assumption, the proposed method uniquely identifies the intervention targets. In the presence of latent confounders, a candidate intervention target set is returned which is a superset of the true intervention targets. The proposed method improves upon the state of the art as the returned candidate set is always a subset of the target set returned by previous work.