Reproduce “Causal Autoregressive Flows” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ilyes Khemakhem, Ricardo Pio Monti, Robert Leech, Aapo Hyvärinen
CAREFL uses a constrained autoregressive normalizing flow to treat variable ordering as a causal order. It keeps that order across affine transformations, allowing causes to shift effects and modulate noise scale, then compares likelihoods for competing directions. On 108 Cause Effect Pairs, it oriented 73% correctly, versus 69% for ANM and RECI; this is benchmark accuracy, not causal certainty.
Key insights
- The inherent autoregressive structure of normalizing flows can define explicit causal orderings, enabling direct construction of Structural Equation Models (SEMs).
- Identifiability can be achieved for a broader class of causal models beyond additive noise, specifically affine models where noise variance depends on the cause.
- Normalizing flows provide a unified computational mechanism for key causal inference tasks due to their exact likelihood estimation and invertible transformations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.