Generative Modeling with Bayesian Sample Inference

Marten Lienen, Marcel Kollovieh, Stephan Günnemann
2/11/2025
cs.LGstat.ML

Official code: martenlienen/bsi (24 stars, MIT, last commit 2025-05-17).

Abstract

We derive a novel generative model from iterative Gaussian posterior inference. By treating the generated sample as an unknown variable, we can formulate the sampling process in the language of Bayesian probability. Our model uses a sequence of prediction and posterior update steps to iteratively narrow down the unknown sample starting from a broad initial belief. In addition to a rigorous theoretical analysis, we establish a connection between our model and diffusion models and show that it includes Bayesian Flow Networks (BFNs) as a special case. In our experiments, we demonstrate that our model improves sample quality on ImageNet32 over both BFNs and the closely related Variational Diffusion Models, while achieving equivalent log-likelihoods on ImageNet32 and ImageNet64. Find our code at https://github.com/martenlienen/bsi.

View on arXivView PDF

Code Implementations(1)

martenlienen/bsiOfficial100%
MIT

Generative Modeling with Bayesian Sample Inference

245Python, Jupyter NotebookFeb 12, 20251 years agoMIT
bayesian-inferencegenerative-model

Similar Papers

Bayesian Flow Networks

87% match1 repo3162023

Generative Uncertainty in Diffusion Models

86% match1 repo162025

Joint Bayesian Inference of Graphical Structure and Parameters with a Single Generative Flow Network

85% match1 repo172023

Posterior Mean Matching: Generative Modeling through Online Bayesian Inference

85% match2024

Outsourced diffusion sampling: Efficient posterior inference in latent spaces of generative models

85% match2026

Inflationary Flows: Calibrated Bayesian Inference with Diffusion-Based Models

84% match1 repo12024

Cite this paper

@article{lienen2025generative,
  title  = {Generative Modeling with Bayesian Sample Inference},
  author = {Marten Lienen and Marcel Kollovieh and Stephan Günnemann},
  year   = {2025},
  eprint = {2502.07580},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2502.07580v3}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2502.07580.svg)](https://codeofpaper.com/paper/2502.07580)

Discussion