Amortized Inference

Definition

A thinking tool distinguishing slow, exact (Bayesian/Monte Carlo) inference — sampling over candidate causes to compute a posterior — from amortized inference, where a trained feedforward map directly outputs the most likely answer, trading exactness for speed. Test-time compute reverts toward full inference; distillation re-amortizes discovered capabilities.

Cast within

No topic-region cast yet — this appears once Amortized Inference's compiled claims are aligned into a topic region's argument tree.

Claims mentioning Amortized Inference (6)

Your notes

Loading notes...