Robustness to graph misspecification

Characterize how missing, spurious, or incorrectly oriented causal edges degrade scope reduction and observational-prior quality, and develop Causal Bayesian Optimization methods that are robust to bounded graph errors.

Background

CBO commonly uses a supplied causal graph to prune intervention scopes and construct observationally informed priors. Errors in the graph can therefore affect both the search space and the surrogate model.

The benchmark includes controlled graph perturbation tests, but the paper states that a systematic characterization of degradation and robust algorithm design remain unresolved.

References

The impact of such errors on CBO performance has been largely unexplored. Future work should characterize how graph misspecification degrades scope reduction and prior quality and develop methods that are robust to bounded graph errors.

— Causal Bayesian Optimization: Foundations, Methods, and Applications  (2609.24112 - Huang et al., 21 Sep 2026) in Section 6.1, paragraph “Graph misspecification”

The POMIS-based scope reduction in CBO eliminates dominated scopes using observational evidence, but the sample complexity of this elimination, i.e., how many observational samples are needed to reliably identify the correct POMIS, has not been analyzed.

— Causal Bayesian Optimization: Foundations, Methods, and Applications  (2609.24112 - Huang et al., 21 Sep 2026) in Section 6.5, paragraph “Sample complexity of scope reduction”

The graph-misspecification and omitted-variable stress tests in Appendix~\ref{appendix:robustness_stress_tests} are a first step toward the robustness evaluation needed for deployment, but comprehensive robustness across the dimensions listed above remains an open evaluation dimension.

— Causal Bayesian Optimization: Foundations, Methods, and Applications  (2609.24112 - Huang et al., 21 Sep 2026) in Conclusion, paragraph “Broader impact and deployment caution”