Summary
Sometimes, different explanations about how things cause each other can look the same when we just watch but disagree when we try changing things. The authors created a method called Interventional Separation Selection (ISS) that asks targeted questions by making specific changes to find where models disagree and eliminate those that don't match what really happens. This process keeps going until no affordable change can tell the remaining models apart, providing a reliable certificate that the models agree on how the system works under those changes. This helps check the understanding of models, like neural networks, about cause and effect beyond what data alone can show.
What this means in practice
- •For machine learning engineers: Verify that different trained models agree on intervention effects within a cost budget using limited active queries to the system.
- •For autonomous system developers: Certify that system models align on cause-effect predictions when applying interventions, improving trust in decision-making modules.
Authors
Sourena Khanzadeh, Daniel Platnick, Marjan Alirezaie, Hossein Rahnama
Abstract
Observationally equivalent causal models can still disagree about what happens under intervention, because interventions create inputs that never occur in observational data. We introduce Interventional Separation Selection (ISS), which repeatedly queries the true system with an admissible intervention on which the surviving candidate models disagree, discards the candidates the outcome contradicts, and stops once no intervention within a cost bound separates the survivors. If the true system is among the candidates, this stopping condition certifies that every survivor agrees with it on every admissible intervention within the bound, a guarantee that no observational learner can give, however much data it sees. The stopping condition depends only on the survivors, so it can be checked without knowing the truth. For continuous variables the candidates form an infinite version space, and mixed-integer linear programs decide the stopping condition exactly over all of it, with agreement holding up to a tolerance. On a three-digit colored MNIST causal abstraction task in which ink hue tracks digit size, plain convolutional networks trained on examples reach zero held-out error, yet disagree with shape-based labels on 26% of single-digit edits, as often as hue-based labels do. Auditing the causal abstractions of networks observed only on such images, ISS certifies what each network perceives with 13.6 interventions per image on average, and each certificate, checked against every admissible intervention, holds whenever the network's true abstraction is among the candidates. When a network bypasses a unit that every candidate abstraction relies on, certificates covering interventions on that unit can be silently void, and twenty random validation interventions refute 69% of them.