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 Regularized Unbalanced Optimal Transport: Single Network, Least Action

Authors: Yuhao Sun, Zhenyi Zhang, Zihan Wang, Tiejun Li, Peijie Zhou

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validated the effectiveness of Var-RUOT on both simulated data and real single-cell datasets. Compared with existing algorithms, Var-RUOT can find solutions with lower action while exhibiting faster convergence and improved training stability. Our code is available at https://github.com/Zeroo Vector/Var RUOT.
Researcher Affiliation Academia 1Center for Machine Learning Research, Peking University. 2LMAM and School of Mathematical Sciences, Peking University. 3Center for Quantitative Biology, Peking University. 4NELBDA, Peking University. 5AI for Science Institute, Beijing. Emails: EMAIL, EMAIL
Pseudocode Yes Algorithm 1 Training Var-RUOT
Open Source Code Yes Our code is available at https://github.com/Zeroo Vector/Var RUOT.
Open Datasets Yes In the experiments presented below, unless the use of the modified metric is explicitly stated, we utilize the standard WFR metric, namely, ψ1(g) = 1. To evaluate the ability of Var-RUOT to capture the minimum-action trajectory, we first conducted experiments on a three-gene simulation dataset (Zhang et al. 2025a). ... on an Epithelial Mesenchymal Transition (EMT) dataset(Sha et al. 2024; Cook and Vanderhyden 2020). ... on the Mouse Blood Hematopoiesis dataset (Weinreb et al. 2020; Sha et al. 2024). ... The fourth dataset is from the Neur IPS Challenge (Luecken et al. 2021).
Dataset Splits Yes For a dataset with n time points, we perform n experiments. In each experiment, one time point is removed from the n time points, and the model is trained using the remaining time points. Afterwards, we compute the W1 and W2 distances between the predicted distribution and the true distribution at the missing time point. ... For this, we extended the three-gene simulation dataset to nine time points. The initial five time points (t = 0, 1, 2, 3, 4) were designated for training, while the final four (t = 5, 6, 7, 8) were reserved for evaluating extrapolation performance.
Hardware Specification Yes We used NVIDIA A100 GPUs (with 40G memory) and 128-core CPUs to conduct the experiments described in this paper.
Software Dependencies No We compute these two metrics using the emd function from the pot library. ... We use the Samples Loss class from the geomloss library to compute the OT loss at each epoch during the training process for each method ... visualized the vector fields u(x, t) on the reduced dimensions using the scvelo library.
Experiment Setup Yes The neural network used to fit λ(x, t) is a fully connected network augmented with layer normalization and residual connections. It consists of 2 hidden layers, each with 512 dimensions. In our algorithm, the main hyperparameters that need tuning include the penalty coefficient α for growth in the action, and the weights γHJB and γAction for the two regularization losses, LHJB and LAction, respectively. ... The parameters used in each case are listed in Table 6. ... Algorithm 1 Training Var-RUOT Require: Datasets D1, . . . , DK, batch size N, training epochs NEpoch, initialized network λθ(x, t). ... setting the integration step size to approximately t = 0.1.