Sequential multi-stage intervention policies

Extend Causal Bayesian Optimization to sequential multi-stage intervention policies in which each intervention depends on outcomes observed at previous stages.

Background

Functional and contextual CBO methods address single-step or few-step policy selection within an explicit causal model. They do not generally solve sequential decision problems in which actions are adapted across multiple stages.

The paper identifies this extension as a bridge between CBO and causal reinforcement learning and leaves it unresolved.

References

The main limitation relative to RL is the assumption of a single-step (or few-step) decision problem; extending CBO to sequential multi-stage intervention policies remains an open challenge.

— Causal Bayesian Optimization: Foundations, Methods, and Applications  (2609.24112 - Huang et al., 21 Sep 2026) in Section 5, paragraph “Policy search and reinforcement learning”