Sanity Checking Causal Representation Learning on a Simple Real-World System
This work evaluates methods for causal representation learning on a simple, real-world system that satisfies the basic problem setup of CRL, and finds that most methods already fail on this synthetic ablation despite its simple data-generating process.