Learning graph generative models over latent spaces has received less attention compared to models that operate on the original data space and has so far demonstrated lacklustre performance. We present GLAD a latent space graph generative model. Unlike most previous latent space graph generative models, GLAD operates on a discrete latent space that preserves to a significant extent the discrete nature of the graph structures making no unnatural assumptions such as latent space continuity. We learn the prior of our discrete latent space by adapting diffusion bridges to its structure. By operating over an appropriately constructed latent space we avoid relying on decompositions that are often used in models that operate in the original data space. We present experiments on a series of graph benchmark datasets which clearly show the superiority of the discrete latent space and obtain state of the art graph generative performance, making GLAD the first latent space graph generative model with competitive performance. Our source code is published at: \url{https://github.com/v18nguye/GLAD}.

翻译：相较于直接作用在原始数据空间的模型，潜空间图生成模型的学习受到的关注较少，且迄今表现欠佳。本文提出GLAD——一种潜空间图生成模型。与以往大多数潜空间图生成模型不同，GLAD在离散潜空间上运行，该空间在很大程度上保留了图结构的离散特性，无需做出潜空间连续性等不自然的假设。通过将扩散桥适应于离散潜空间的结构，我们学习该空间的先验分布。通过作用在合理构建的潜空间上，我们避免了依赖于常被用于原始数据空间模型的分解方法。我们在系列图基准数据集上进行了实验，结果清晰展现了离散潜空间的优越性，并取得了最先进的图生成性能，使GLAD成为首个具有竞争力表现的潜空间图生成模型。我们的源代码发布于：\url{https://github.com/v18nguye/GLAD}。