# Piecewise deterministic generative models

Source: [Piecewise deterministic generative models](<https://www.sota2.com/research/paper/68fa75891e209131cea686d5>)

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

Data updated: 2026-01-02T07:57:13.273Z

Authors: Andrea Bertazzi, Dario Shariatian, Umut Simsekli, Eric Moulines, Alain Durmus

Published: 2024-07-28T09:53:02.000Z

## Abstract

We introduce a novel class of generative models based on piecewise deterministic Markov processes (PDMPs), a family of non-diffusive stochastic processes consisting of deterministic motion and random jumps at random times. Similarly to diffusions, such Markov processes admit time reversals that turn out to be PDMPs as well. We apply this observation to three PDMPs considered in the literature: the Zig-Zag process, Bouncy Particle Sampler, and Randomised Hamiltonian Monte Carlo. For these three particular instances, we show that the jump rates and kernels of the corresponding time reversals admit explicit expressions depending on some conditional densities of the PDMP under consideration before and after a jump. Based on these results, we propose efficient training procedures to learn these characteristics and consider methods to approximately simulate the reverse process. Finally, we provide bounds in the total variation distance between the data distribution and the resulting distribution of our model in the case where the base distribution is the standard $d$-dimensional Gaussian distribution. Promising numerical simulations support further investigations into this class of models.

## 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.

- [2D Synthetic Data Generation on Checkerboard (test)](<https://www.sota2.com/research/sota/2d-synthetic-data-generation-on-checkerboard-test.md>): 2D Synthetic Data Generation; checkerboard (test); MMD: 0.81.

- [2D Synthetic Data Generation on Fractal tree (test)](<https://www.sota2.com/research/sota/2d-synthetic-data-generation-on-fractal-tree-test.md>): 2D Synthetic Data Generation; Fractal tree (test); MMD: 1.12.

- [2D Synthetic Data Generation on Gaussian grid (test)](<https://www.sota2.com/research/sota/2d-synthetic-data-generation-on-gaussian-grid-test.md>): 2D Synthetic Data Generation; Gaussian grid (test); MMD: 4.01.

- [2D Synthetic Data Generation on Olympic rings (test)](<https://www.sota2.com/research/sota/2d-synthetic-data-generation-on-olympic-rings-test.md>): 2D Synthetic Data Generation; Olympic rings (test); MMD: 1.43.

- [2D Synthetic Data Generation on Rose (test)](<https://www.sota2.com/research/sota/2d-synthetic-data-generation-on-rose-test.md>): 2D Synthetic Data Generation; Rose (test); MMD: 0.9.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2407.19448>)

[PDF](<https://arxiv.org/pdf/2407.19448.pdf>)
