View a PDF of the paper titled Simplifying debiased inference through automated differentiation and probabilistic programming, by Alex Luedtke
Summary:We introduce an algorithm that simplifies the development of environment friendly estimators, making them accessible to a broader viewers. ‘Dimple’ takes as enter pc code representing a parameter of curiosity and outputs an environment friendly estimator. Not like commonplace approaches, it doesn’t require customers to derive a practical by-product often called the environment friendly affect perform. Dimple avoids this job by making use of automated differentiation to the statistical practical of curiosity. Doing so requires expressing this practical as a composition of primitives satisfying a novel differentiability situation. Dimple additionally makes use of this composition to find out the nuisances it should estimate. In software program, primitives might be applied independently of each other and reused throughout completely different estimation issues. We offer a proof-of-concept Python implementation and showcase by way of examples the way it permits customers to go from parameter specification to environment friendly estimation with just some traces of code.
Submission historical past
From: Alex Luedtke [view email]
[v1]
Tue, 14 Could 2024 14:56:54 UTC (100 KB)
[v2]
Thu, 20 Jun 2024 23:30:35 UTC (145 KB)