Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Piecewise deterministic generative models
Authors: Andrea Bertazzi, Dario Shariatian, Umut Simsekli, Eric Moulines, Alain Durmus
NeurIPS 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Promising numerical simulations support further investigations into this class of models. In Section 4 we test our models on simple toy distributions. |
| Researcher Affiliation | Academia | 1 École Polytechnique, Institut Polytechnique de Paris 2 INRIA, CNRS, Ecole Normale Supérieure, PSL Research University 3 MBZUAI |
| Pseudocode | Yes | Algorithm 1: Pseudo-code for the simulation of a homogeneous PDMP |
| Open Source Code | Yes | we provide all the necessary codes to reproduce our experiments. |
| Open Datasets | Yes | We consider the task of generating handwritten digits training the ZZP on the MNIST dataset. |
| Dataset Splits | No | The paper specifies training and test sample counts but does not explicitly mention a validation set or its split details. |
| Hardware Specification | Yes | We run our experiments on 50 Cascade Lake Intel Xeon 5218 16 cores, 2.4GHz. |
| Software Dependencies | No | The paper mentions software like PyTorch, zuko, and Adam by name but does not provide specific version numbers for these dependencies. |
| Experiment Setup | Yes | For each forward PDMP, we take a time horizon Tf equal to 5, and set the refreshment rate λr to 1. The optimiser is Adam [Kingma and Ba, 2015] with learning rate 5e-4 for all neural networks. We use a batch size of 4096 and train our model for 25000 steps. |