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 [1].
Field Matching: an Electrostatic Paradigm to Generate and Transfer Data
Authors: Alexander Kolesov, S. I. Manukhov, Vladimir Vladimirovich Palyulin, Alexander Korotin
ICML 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we demonstrate the proof-of-concept experiments with our proposed EFM method. We show a 2-dimensional illustrative experiment ( 4.1), image-to-image translation experiment ( 4.2) and image generation experiment ( 4.3) with the colored MNIST and CIFAR-10 datasets. |
| Researcher Affiliation | Academia | 1Skolkovo Institute of Science and Technology, Moscow, Russia 2Artificial Intelligence Research Institute, Moscow, Russia 3Lomonosov Moscow State University, Faculty of Physics, Moscow, Russia. Correspondence to: Alexander Kolesov <EMAIL>. |
| Pseudocode | Yes | Algorithm 1 EFM Training Algorithm 2 EFM Sampling |
| Open Source Code | Yes | Our code is available at https://github.com/justkolesov/FieldMatching |
| Open Datasets | Yes | We show a 2-dimensional illustrative experiment ( 4.1), image-to-image translation experiment ( 4.2) and image generation experiment ( 4.3) with the colored MNIST and CIFAR-10 datasets. |
| Dataset Splits | No | The paper uses well-known datasets like MNIST and CIFAR-10 but does not explicitly describe the training/test/validation splits used for its experiments, nor does it refer to specific standard splits from citations. |
| Hardware Specification | Yes | Evaluation of the training time for our solver on the image s experiments (see 4.2 and 4.3)takes less than 10 hours on a single GPU GTX 1080ti (11 GB VRAM). |
| Software Dependencies | No | The paper mentions using Adam optimizer and RK45 ODE solver (from scipy), and refers to source codes of PFGM and Flow Matching, but it does not specify version numbers for these software components or any other key libraries/frameworks. |
| Experiment Setup | Yes | We aggregate the hyper-parameters of our Algorithm 1 for different experiments in the Table 1. ... We use Exponential Moving Averaging (EMA) technique with the ema rate decay equals to 0.99 . As for the optimization procedure, we use Adam optimizer (Kingma & Ba, 2015) with the learning rate λ = 2e 4 and weight decay equals to 1e-4. ... We found the following values of hyper parameters are appropriate for us: γ = 5, t = 0.3, ε = 1e 3, see (Xu et al., 2022) for details. |