New arXiv paper: density ratio estimation via Stein displacement fields
_onionesque · x · 2026-10-12
Song Liu's new arXiv paper models the density ratio between target and base distributions by parametrizing it through a displacement field: the log-ratio equals the base's Stein operator applied to the field, up to a normalizing constant, yielding both statistical and dynamical views of distribution shift from a single convex optimization. Iterating the estimate-and-move step gives two algorithms — push-forward, which corrects a pretrained sampler without retraining, and pull-back, which fits a transformation layer by layer. Applications include distribution shift in simulation-based inference and nonlinear ICA.
More from Research
- OpenAI raises the plane's chromatic number lower bound to 6, Lean-formalized — fortnow · 2026-10-12
- Snorkel AI Expands Open Benchmark Grants 10x to $30M, Adds Red Team — ajratner · 2026-10-12
- How long is the coast of a neuron? Point-neuron abstraction may hide capacity — neurovium · 2026-10-12
- Community shrinks OpenAI Problem #190 obstruction from 66x66 to 3x4, verified in Lean — BorisMPower · 2026-10-12
- Autoresearch for LLM pretraining needs humans in the loop; stacked changes are the killer — menhguin · 2026-10-12
- COMODO distills video intelligence into IMU sensors for camera-free egocentric AI glasses — flosalim · 2026-10-12