Computational Identifiability
A new paper introduces the concept of "computational identifiability" as a practical alternative to theoretical identifiability in machine learning. This framework defines identifiability based on the success of a finite computational search procedure for an empirical estimator, rather than relying on idealized asymptotic conditions. The approach allows for answering fine-grained identification questions concerning small sample sizes, ambiguous graphical criteria, and mixed observational-interventional data. The authors provide experimental demonstrations and make their code available. AI
IMPACT Introduces a new framework for addressing practical identifiability challenges in machine learning models.
- Hugging Face
- arXiv
- machine learning
- Computational Identifiability
- causal graph
- estimator
- Empirical estimators of gamma fits to tracer-dilution curves and their technical basis and practical scope
- finite samples
- mixed observational-interventional data
- Counterfactual Data Augmentation for Debiased Coupon Recommendations Based on Potential Knowledge