# Score-Based Density Estimation from Pairwise Comparisons

Source: [Score-Based Density Estimation from Pairwise Comparisons](<https://www.sota2.com/research/paper/68fa7bff1e209131cea8bc60>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-03-27T05:00:04.193Z

Authors: Petrus Mikkola, Luigi Acerbi, Arto Klami

Published: 2025-10-10T08:49:24.000Z

## Abstract

We study density estimation from pairwise comparisons, motivated by expert knowledge elicitation and learning from human feedback. We relate the unobserved target density to a tempered winner density (marginal density of preferred choices), learning the winner's score via score-matching. This allows estimating the target by \`de-tempering' the estimated winner density's score. We prove that the score vectors of the belief and the winner density are collinear, linked by a position-dependent tempering field. We give analytical formulas for this field and propose an estimator for it under the Bradley-Terry model. Using a diffusion model trained on tempered samples generated via score-scaled annealed Langevin dynamics, we can learn complex multivariate belief densities of simulated experts, from only hundreds to thousands of pairwise comparisons.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Density estimation from pairwise comparisons on Onemoon2D](<https://www.sota2.com/research/sota/density-estimation-from-pairwise-comparisons-on-onemoon2d.md>): Density estimation from pairwise comparisons; Onemoon2D; Wasserstein Distance: 0.37.

- [Density estimation from pairwise comparisons on Twomoons2D](<https://www.sota2.com/research/sota/density-estimation-from-pairwise-comparisons-on-twomoons2d.md>): Density estimation from pairwise comparisons; Twomoons2D; Wasserstein Distance: 0.44.

- [Density estimation from pairwise comparisons on Ring2D](<https://www.sota2.com/research/sota/density-estimation-from-pairwise-comparisons-on-ring2d.md>): Density estimation from pairwise comparisons; Ring2D; Wasserstein Distance: 0.39.

- [Density estimation from pairwise comparisons on Gaussian4D](<https://www.sota2.com/research/sota/density-estimation-from-pairwise-comparisons-on-gaussian4d.md>): Density estimation from pairwise comparisons; Gaussian4D; Wasserstein Distance: 1.4.

- [Density estimation from pairwise comparisons on Mixturegaussians4D](<https://www.sota2.com/research/sota/density-estimation-from-pairwise-comparisons-on-mixturegaussians4d.md>): Density estimation from pairwise comparisons; Mixturegaussians4D; Wasserstein Distance: 1.09.

- [Density estimation from pairwise comparisons on Stargaussian6D](<https://www.sota2.com/research/sota/density-estimation-from-pairwise-comparisons-on-stargaussian6d.md>): Density estimation from pairwise comparisons; Stargaussian6D; Wasserstein Distance: 1.28.

- [Density estimation from pairwise comparisons on Mixturegaussians 10D](<https://www.sota2.com/research/sota/density-estimation-from-pairwise-comparisons-on-mixturegaussians-10d.md>): Density estimation from pairwise comparisons; Mixturegaussians 10D; Wasserstein Distance: 1.1.

- [Density estimation from pairwise comparisons on Gaussian16D](<https://www.sota2.com/research/sota/density-estimation-from-pairwise-comparisons-on-gaussian16d.md>): Density estimation from pairwise comparisons; Gaussian16D; Wasserstein Distance: 5.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2510.09146>)

[PDF](<https://arxiv.org/pdf/2510.09146.pdf>)
