Density Ratio Estimation with Stein Displacement Fields (arXiv 2610.12437)

Forum topic · 小凯 · 2026-10-10

Summary

This paper, authored by Song Liu (arXiv 2610.12437), proposes estimating the density ratio between a target and a base distribution by parametrizing it through a displacement field acting on the base distribution. The log-ratio is modeled as minus the Stein operator of the base applied to the field, up to a normalizing constant, yielding both a statistical and a dynamical description of distribution shift via a single convex optimization problem. Iterating this estimate-and-move step produces two inference algorithms: push-forward, which moves a model and corrects pre-trained samplers without retraining, and pull-back, which moves data toward the base distribution while fitting transformation models layer by layer. Applications to distribution shift in simulation-based inference and nonlinear independent component analysis demonstrate the method's benefits and limitations.

Density Ratio Estimation with Stein Displacement Fields

Field: Machine Learning
Author: Song Liu
Posted: 2026-10-08
arXiv: 2610.12437

Overview

Density ratios quantify distribution shift from a probability-mass point of view, whereas displacement fields describe, from a dynamical point of view, how one distribution is transported onto another. Although both offer complementary insights, they are usually estimated separately, and converting one into the other requires post-processing.

Method

The paper estimates the density ratio between a target and a base distribution by parametrizing it through a displacement field acting on the base: the log-ratio is modeled as minus the Stein operator of the base applied to the field, up to a normalizing constant. This gives both statistical and dynamical descriptions of the distribution shift through a single convex optimization problem.

Algorithms

Iterating this estimate-and-move step yields two inference algorithms:

Push-forward: moves the model and corrects pre-trained samplers without retraining.
Pull-back: moves the data toward the base distribution while fitting transformation models layer by layer.

Experiments

Applications to distribution shift in simulation-based inference and nonlinear independent component analysis showcase the method's benefits and limitations.

--- *Auto-collected on 2026-10-10*