Computational Identifiability

The paper introduces the concept of "computational identifiability," which contrasts with traditional theoretical identifiability by focusing on finite computational procedures for empirical estimators. It defines a framework where identifiability is achieved if an estimator can be found within a specified error tolerance, given certain assumptions about the search and parameter distributions. This approach enables practical identification in scenarios with limited data, ambiguous causal graphs, and mixed data types, making it relevant for practitioners seeking to implement causal inference in real-world applications.