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..
Variational Flow Matching for Graph Generation
Authors: Floor Eijkelboom, Grigory Bartosh, Christian Andersson Naesseth, Max Welling, Jan-Willem van de Meent
NeurIPS 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate Cat Flow on one abstract graph generation task and two molecular generation tasks. In all cases, Cat Flow exceeds or matches performance of the current state-of-the-art. We evaluate Cat Flow in three sets of experiments. |
| Researcher Affiliation | Collaboration | Floor Eijkelboom Uv A-Bosch Delta Lab University of Amsterdam Grigory Bartosh AMLab University of Amsterdam Christian A. Naesseth Uv A-Bosch Delta Lab University of Amsterdam Max Welling Uv A-Bosch Delta Lab University of Amsterdam Jan-Willem van de Meent Uv A-Bosch Delta Lab University of Amsterdam |
| Pseudocode | Yes | Algorithm 1 Variational Flow Matching; Algorithm 2 Categorical Variational Flow Matching (Cat Flow); Algorithm 3 Gaussian Variational Flow Matching (all in Appendix B). |
| Open Source Code | Yes | We provide all code to the experiments. Moreover, all data is publically available, and will be provided in the repository, including data processing. |
| Open Datasets | Yes | We evaluate Cat Flow on two popular molecular generation benchmarks: QM9 [40] and ZINC250k [19]. |
| Dataset Splits | Yes | Furthermore, all data splits are kept the same as in [20] |
| Hardware Specification | Yes | All experiments were run on a single NVIDIA RTX 6000 and took about a day to run. |
| Software Dependencies | No | The paper mentions 'Adam W' as an optimizer and 'RDKit' but does not provide specific version numbers for these or other software dependencies. |
| Experiment Setup | Yes | We report the hyperparameters here: Optimizer Adam W, Scheduler Cosine Annealing, Learning Rate 2e-4, Weight Decay 1e-12, EMA 0.999 (Appendix D.2, Table 3). |