Mixed intervention types in one optimization loop

Construct a Causal Bayesian Optimization method that handles arbitrary mixtures of hard, soft, and stochastic interventions within a single optimization loop.

Background

Practical systems may combine discrete or clamping actions, mechanism modifications, and stochastic treatment or assignment policies. Existing CBO variants generally specialize in one or a limited subset of these intervention representations.

The paper explicitly identifies the absence of a unified method for arbitrary mixtures as an unresolved modeling problem.

References

No existing CBO method handles arbitrary mixtures of these types within a single optimization loop.

— Causal Bayesian Optimization: Foundations, Methods, and Applications  (2609.24112 - Huang et al., 21 Sep 2026) in Section 6.3, paragraph “Mixed intervention types”