Amortized Inference for Causal Structure Learning
What this work does
Amortizes graph inference with a permutation-aware neural model trained on sampled graph/mechanism/data distributions under a forward-KL objective.
Connection to discovery
A useful model of learning an inference procedure across worlds; separates amortized causal inference from learning a policy to gather data.
Evidence
Inspected the simulation-trained objective, optional acyclicity constraint and graph/mechanism/noise distribution-shift evaluations.
Scope and limitations
The simulator distribution acts as a prior. Amortization and good synthetic OOD scores do not overcome unobserved confounding or identification failures outside its assumptions.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.