Score-based Generative Modeling Secretly Minimizes the Wasserstein Distance

Authors: Dohyun Kwon, Ying Fan, Kangwook Lee

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our numerical experiments support our findings. By analyzing our upper bounds, we provide a few techniques to obtain tighter upper bounds.
Researcher Affiliation Academia Dohyun Kwon, Ying Fan, Kangwook Lee University of Wisconsin-Madison
Pseudocode No No pseudocode or algorithm block was found in the paper.
Open Source Code Yes Code is available at https://github.com/UW-Madison-Lee-Lab/score-wasserstein.
Open Datasets Yes Here we adopt three 2D datasets for simulation: One cluster Gaussian N(0, 0.1I), two moons in [28], and four clusters Gaussian mixture N(( 0.5, 0.5) , 0.01I) with equal weights for each cluster.
Dataset Splits No The paper does not explicitly provide training/validation/test dataset splits with percentages, sample counts, or references to predefined splits for the datasets used.
Hardware Specification No No specific hardware details (e.g., GPU models, CPU types, or memory) used for running the experiments are mentioned.
Software Dependencies No The paper mentions software like Adam W, POT, and scikit-learn, but does not specify their version numbers.
Experiment Setup Yes We use a 4-layer neural network as the score matching model, with Re LU nonlinearity and skip-connection at the final output. Each layer is composed of a linear layer with 64 hidden neurons and an embedding layer for 10 timesteps. For optimizer, we use Adam W [22] with learning rate = 0.001 and weight decay coefficient 0.01. For loss function, we use JDSM with λ(t) = g(t)2 and batch size = 128.